Fullscreen HTML5Med Res (??MB)HTML4

HUMANOID & AN Agility Digit www.agilityrobotics.com Boston Dynamics Atlas https://bostondynamics.com/atlas Unitree H1 www.unitree.com/h1 Tesla Optimus www.tesla.com/en_eu/AI Like many ideas that started as science fiction, humanoid and android robots are now a reality. They have not yet been perfected – but they are here. Last month we covered the tech; and now we showcase some of the most interesting robots. W e will now look at some historical humanoid robots, followed by those that are under active development or are available commercially. Leonardo da Vinci’s mechanical knight Around 1495, Leonardo conceived a “mechanical knight” that could perform actions such as moving its arms, neck & jaw, raising its visor, sitting and standing (Fig.18). It was operated with gears and pulleys but obviously had no electronic intelligence. The significance of this robot is that it is thought to be the first demonstration of the ability to mimic human-like actions by mechanical means. Leonardo’s sketches for this machine were rediscovered in the 1950s by Carlo 16 Silicon Chip Pedretti and, in 1993, Mark Rosheim collaborated with him to reconstruct the robot. As designed by Leonardo, it has three degrees of freedom in its legs and four in its arms. Pedretti described it as “the first articulated humanoid robot in the history of Western Civilisation”. There is a BBC audio presentation about the robot at www.bbc. co.uk/sounds/play/m0004mf2 Elecktro Elektro was an early humanoid robot built by Westinghouse in 1937 for the 1939/1940 World’s Fair in New York. It was extremely impressive and popular at the time, and it was featured in daily shows at the fair – see Fig.19. It was 2.1m tall, weighed 120kg, Australia's electronics magazine could perform 26 actions and had 700 words stored on eight 78 RPM records in its chest cavity. The 26 actions were hard-wired in a fixed sequence so that in a show, the same script from the handler resulting in the same sequence of actions every time. The robot gave the illusion of being voice-controlled, but electronics of the era were not advanced enough to perform voice recognition. The robot responded to the controller’s voice, but only to the rhythm. For example, speaking one word might correspond to one impulse to close a relay. Two words would correspond to two impulses which might perform another electro-mechanical action. It could also recognise red or green colours with a photocell, and state siliconchip.com.au NDROID ROBOTS Part 2: by Dr David Maddison, VK3DSM Figure 02 www.figure.ai 1X NEO Gamma www.1x.tech/neo Apptronik Apollo https://apptronik.com/apollo Fig.18: a reconstruction of Leonardo’s “mechanical knight” robot. Source: https://w. wiki/EotZ Fig.19: the Elektro robot. Source: www.computertimeline. com/timeline/ the-robots-ofwestinghouse the colour. See the video titled “Elektro the Smoking Robot (Odd History)” at https://youtu.be/sxGDdwbcfJg and “The World’s First Celebrity Robot” at https://youtu.be/dwBLOluOiUY The system of voice control had its origins in Westinghouse engineer Roy James Wensley, who was issued a patent in 1929 for a “supervisory control system” that enabled the control siliconchip.com.au of machinery over telephone or radio connection by “voice”. It wasn’t truly voice recognition, but a series of tones generated with pitch pipes or tuning forks at 600, 900 and 1400Hz. These activated relays via tuned circuits, enabling equipment such as at a power plant or telephone exchange to be remotely controlled. Australia's electronics magazine Booster Robotics T1 www.boosterobotics.com/robots/ The system was demonstrated with a robot prior to Elektro called Herbert Televox. Presumably, this system was modified for Elektro to use impulses instead of tones. This system of tone control became the basis of the DTMF (dual-tone multi-frequency) signalling used in traditional telephone exchanges. Current humanoid and android robots We do not have space to include a description of every available humanoid robot, so have selected the most interesting ones. 1X Technologies www.1x.tech The Neo Gamma is a humanoid robot under development by a Norwegian AI and robotics company, intended for domestic service (Fig.20). Neo Gamma uses artificial intelligence and is trained on household tasks using human motion-capture data. It December 2025  17 Fig.20: the Neo Gamma from 1X Technologies performs domestic duties. Source: www. therobotreport.com/1xbuilt-humanoid-neogamma-better-fit-home Fig.21: Australian Abi companion robots. Source: www.dromeda. com.au/product is covered in a knitted fabric to give it a friendly appearance. EVE is another humanoid robot from 1X. Abi www.dromeda.com.au Abi is an Australian humanoid robot, made in Melbourne – see Fig.21. It is intended as a companion robot with “playful features and infinite empathy”, for use in nursing homes and similar care facilities. It uses advanced AI and machine learning to recognise faces, understand and express emotions and remember conversations from days or months ago. It interacts with each resident based on their personal cues, including speaking their preferred language. Abi is fluent in 90 languages and can participate in small group activities such as singing, dancing, games and conversation. Actroids www.kokoro-dreams.co.jp The Actroid-DER series of android robots is designed with strong human likeness by Osaka University and manufactured by the Kokoro Company Ltd. Actroids were first displayed in 2003, so they are relatively old, and various models are available to rent. A model from the Actroid-DER series is shown here. It performs simple functions like blinking and speaking using AI. Later models have 47 actuators, which are pneumatically driven. For more information, see the video at https://youtu.be/l8qHXdKF300 AgiBot www.agibot.com This Chinese company has developed the AI-powered Genie Operator-1 (GO-1) model for quickly training humanoid robots in a variety of tasks – see Fig.23. According to Agibot, the model enables robots to “understand instructions in natural language and perform reasoning, rather than being limited to preprogrammed routines”. AgiBot also produces a variety of robots, such as the AgiBot A2 interactive service robot. The open-source AgiBot X1 robot is documented at www.agibot.com/DOCS/OS Apollo A1 https://apptronik.com Apptronik from Austin, Texas, has developed the Apollo A1 general-­ purpose humanoid robot, intended for jobs in warehouses, manufacturing plants and so on – see Fig.24. As the robot is further developed, its use will be extended to areas like construction, oil and gas extraction, electronics production, retail, home delivery, aged care and others. Apptronik has origins in the development of robots for NASA, such as Valkyrie (described below). Apollo is 172cm tall, has a run time of four hours with a swappable battery pack, weighs 72kg, can carry 25kg, is modular and the torso can even be mounted on a wheeled or stationary platform. It has a chest-mounted display that shows its status. Apollo can be tethered to a power supply to allow continuous operation. It has ‘force control’ systems to limit the amount of mechanical force it can impose to enhance safety around humans. It walks within a defined perimeter, so it does not get too close to people or objects. It immediately pauses when moving objects are detected in its vicinity. Mercedes Benz has an agreement to test Apollo in its manufacturing operations. See the video at https://youtu. be/TfUOg38iXxo ASIMO https://global.honda/en/robotics/ Apart from cars, Honda is famous for one of the earliest humanoid robots, ASIMO (Advanced Step in Innovative Mobility) – see Fig.25. It was retired in 2018. Honda researchers researched the following things with ASIMO: O Moving around while sharing the same space with people. O Performing tasks using its hands. O Interacting with people, including Fig.22: an Actroid-DER series android. Source: www.kokoro-dreams.co.jp/ english/rt_tokutyu/actroid/ Fig.23: AgiBot robots running GO-1 perform some kitchen tasks. Source: https://youtu.be/9dvygD4G93c 18 Silicon Chip Australia's electronics magazine siliconchip.com.au understanding spoken words and controlling movement/behaviour by estimating the intention of nearby people. Over its 20 years of demonstrations, it walked a total of 7907km. Honda continues to develop robotics, but with a focus on developing a variety of robots with specific functions rather than just one general-­ purpose humanoid robot. Fig.24: the Apptronik A1 generalpurpose robot. Source: https:// apptronik.com Digit www.agilityrobotics.com Agility Robotics from Albany, Oregon, USA has its origins in Oregon State University. In 2023, they released the current version of their bipedal robot, digit – see Fig.26. Digit is said to be the world’s first commercially deployed humanoid robot. They are currently in operation at Amazon and GXO Logistics. Amazon has also stated they intend to replace their entire human workforce by 2030. The robots cost around US$250,000 ($380,000) each. See https://youtu.be/NvYsGcQvMw8 and https://youtu.be/MYzRPJ7TaLc AheadForm www.aheadform.com Heads capable of expression are among the most complicated things to build for such robots, apart from the software. Just as motor vehicle manufacturers might contract out specialists to make certain components, the same concept can apply to robots. AheadForm specialises in making robot heads capable of a wide variety of human-like facial expressions, including subtle but important ones such as smirks – see Fig.27. These highly realistic heads are said to avoid the “uncanny valley” in which humanlike heads are seen as not quite realistic enough, and thus disturbing to the viewer. AiMOGA Robot Artificial Intelligence with Multi-Objective Genetic Algorithm was the robot’s original name, but it is now called Mornine. Its job is as an intelligent sales consultant for some Chery car dealerships, with about 220 to be delivered – see Fig.28. The robot is manufactured by Chery. It uses a ‘multimodal sensory model’ to perceive a customer’s gestures and questions by combining sensory data from speech, visual and other data. It then uses an “advanced emotion and personality engine” for personalised interactions. siliconchip.com.au Fig.25: Honda’s ASIMO (Advanced Step in Innovative Mobility) showcased at the Tokyo Motor Show in 2011. Source: https://w.wiki/EpwP Fig.26: a demonstration of Digit robots from Agility Robotics at work in a warehouse. Fig.27: the AheadForm robot head (left) and its creator (right). Source: https://youtu. be/gnyFtUU-TJ8 Fig.28: a Mornine robot at a Chery dealership. Source: https:// motorillustrated.com/ chery-debuts-humanoidrobot-mornine-to-dealerstheir-future-salesreps/153657/ Australia's electronics magazine December 2025  19 It takes advantage of DeepSeek’s AI and CheryGPT’s large language models to understand natural language, give appropriate responses and chat in any of ten languages. The robot can be used for other purposes apart from car sales, such as a bookshop assistant, other sales roles, companionship, shopping guide or caregiver. The robot is 166cm tall and weighs 55kg. Chery intends to launch the robot commercially in 2027. One estimate is that the robots will cost around $89,000 each. Fig.29: Alter3 after being instructed to “take a selfie”. Source: https://arxiv. org/pdf/2312.06571 Fig.30: the Ameca robot. Source: https://engineeredarts.com/gallery/ Fig.31: the Berkeley Humanoid Lite. Fig.32: the Booster T1 robot. Source: www. boosterobotics.com/ robots 20 Alter3 https://tnoinkwms.github.io/ALTER-LLM Alter3 is an experimental humanoid robot from the University of Tokyo that uses the GPT-4 large language model to interpret instructions, then produces separate code to generate the required motions – see Fig.29. There is a video showing the implementation of this code at https://youtu. be/l4d6N_Rf8mk Ameca https://engineeredarts.com A humanoid robot by UK company Engineered Arts (Fig.30), Ameca is designed for interaction with humans; its legs are still under development. Despite that, Ameca has advanced facial expression capabilities, extremely advanced emotional intelligence and conversational capabilities in multiple languages. It runs an in-house-developed operating framework called Tritium; for its language model, it uses OpenAI’s GPT 4.0 or can be controlled by teleoperation. It can watch, listen, learn, track faces in real time, analyse the emotional state of someone speaking to it and respond appropriately. Ameca is available for purchase or rental. See https://youtu.be/b9xFM61KKyc Fig.33: the latest version of Boston Dynamics’ Atlas robot. Source: https://bostondynamics.com/atlas/ Berkeley Humanoid Lite Berkeley Humanoid Lite (https:// lite.berkeley-humanoid.org) is an open-source, customisable 3D-printed humanoid robot – see Fig.31. It is 88cm tall and weighs 16kg. Its developers state that it can be made for around US$5000 ($7650) with readily available components. It can perform several useful tasks, including manipulating a Rubik’s Cube, and it can also be teleoperated. To make it, you need a 3D printer that can produce parts within a 20 × 20 × 20cm workspace. For more details, see https://youtu.be/dIdJGkMDFl4 Australia's electronics magazine siliconchip.com.au Fig.34: Atlas picks and places parts in an experimental situation. Source: https://bostondynamics.com/video/ pick-carry-place-repeat Fig.35: the torso of the Clone android showing anatomical similarity to a human. Source: https://x.com/ clonerobotics Fig.36: the Pudu D9 robot. Source: www.pudurobotics.com/en/products/ d9 Booster T1 www.boosterobotics.com Booster Robotics is a Chinese company that has developed a humanoid robot platform intended for developers to write their own software for – see Fig.32. There is an online manual describing the robot with details for developers at siliconchip.au/link/ ac7h picking and placing a part is shown in Fig.34. D9 www.pudurobotics.com Pudu Robotics is developing the general-purpose D9 robot – see Fig.36 and the video at https://youtu.be/ gd5DdfJX_RM Boston Dynamics Atlas Boston Dynamics (website: https:// bostondynamics.com) is a subsidiary of Hyundai. It was one of the first companies with a functional humanoid robot called Atlas, and there are many impressive videos of its products on YouTube. The latest version of Atlas is shown in Fig.33. Unlike previous versions, which were hydraulically actuated, this one is fully electric. That makes it quieter, less complicated, more compact, with more natural movements and some say less intimidating. Computationally, Atlas uses NVIDIA’s Isaac GR00T framework and the NVIDIA Jetson Thor computing platform, which is specifically designed for humanoid robots. It uses the Blackwell GPU architecture with 2560 CUDA cores, 96 tensor cores and Arm Neoverse V3AE CPUs with 14 cores and 128GB of LPDDR5X memory, giving up to 2070 FP4 teraflops of AI computing power (FP4 is floating point 4-bit operations). Hyundai are testing Boston Dynamics Atlas robots for building electric cars in their Georgia, US manufacturing facilities and plan to roll them out globally. A demonstration of Atlas Clone https://clonerobotics.com Polish company Clone Robotics developed Clone, which they describe as an android. Clone adopts a different approach from other humanoid robots. It seeks to emulate the actual structure of the human body, with anatomically accurate bones, joints, muscles and a nervous system – see Fig.35. It is hydraulically actuated with a 500W pump. It has an unusually large 200 degrees of freedom, 206 bones, 1000 artificial muscle materials (known as myofibres) and 200 sensors. Myofibres are composed of mesh tubes that contain balloons of hydraulic fluid. These are inflated or deflated by hydraulic pressure. The aforementioned pump acts as the ‘heart’ of the system. The idea is based on the McKibben artificial muscle concept first invented in the 1950s. For cooling, it even ‘sweats’ fluid, just like a human. In essence, Clone seeks to mimic the human body. Clone runs on an NVIDIA Jetson Thor inference GPU in the skull running Cybernet, Clone’s visuomotor foundation model. See Clone Robotics’s YouTube channel (www.youtube. com/<at>CloneRobotics). siliconchip.com.au Certis www.certisgroup.com Certis is a Singapore company that is using an Agibot humanoid robot to study potential applications for humanoid robots in security and integrated facilities management. Australia's electronics magazine EveR-4 android The EveR-4 is an android that was developed at the Korea Institute of Industrial Technology and was exhibited at RoboWorld 2011 in Seoul. It is one of a series of four such androids. It can exhibit a wide range of facial expressions and has over thirty motors actuating its face. The EverR-4 has been investigated for use in jobs such as a medical receptionist, where the robot was positively received by patients. See the videos at https://youtu.be/OvsZqPcnNIE and https://youtu.be/b2GtBAPG1ho (EveR-4 as a receptionist). FALCON FALCON is an experimental robot control framework from Carnegie Mellon University, designed to train humanoid robots for a complex tasks called “force-adaptive loco-manipulation”. This is walking or standing and using their arms at the same time, while applying strong, precise force. Examples include pushing a wheelbarrow, opening a door or engaging in a tug -of-war with another robot. Two AI agents are used to accomplish such tasks; one for the legs and the other December 2025  21 for the arms – they communicate with each other. The systems is tested on humanoid platforms from Unitree and Booster Robotics. See https://youtu.be/ OfsvJ5-Fyzg Figure AI www.figure.ai An American robotics company, its investors include OpenAI, NVIDIA and Jeff Bezos. It is aiming to build robots that learn and reason like humans. Their most advanced robot is Figure 02 (shown in Fig.37), which runs the Helix generalist vision language action (VLA) model. The aim of Figure AI is to enable robots to work in an unstructured home environment with thousands of objects it has never encountered before, and to reason how to deal with them without prior programming or demonstrations. Helix can also run on two robots simultaneously, enabling them to cooperatively solve problems and share learning. This is shown by putting groceries away in the video at https://youtu.be/Z3yQHYNXPws Figure 02 with Helix is also intended for industrial use, where it will be introduced first, before being brought to the domestic market. An industrial environment is much more structured than a home environment, and should be easier to operate in. According to Figure AI: VLM (vision language model) backbones are general, but not fast, while robot visuomotor policies are fast but not general. Helix resolves this trade-off through two complementary systems, trained end-to-end to communicate: O System 2 (S2) is an onboard internet-­pretrained VLM operating at 7-9Hz for scene understanding and language comprehension, enabling broad generalisation across objects and contexts. O System 1 (S1) is a fast reactive visuomotor policy that translates the latent semantic representations produced by S2 into precise continuous robot actions at 200Hz. Fig.37: Figure AI’s Figure 02 robot uses a VLA model for control. Source: www.figure.ai 22 Silicon Chip This decoupled architecture allows each system to operate at its optimal timescale. S2 can “think slow” about high-level goals, while S1 can “think fast” to execute and adjust actions in real-time. For example, during collaborative behaviour, S1 quickly adapts to the changing motions of a partner robot while maintaining S2’s semantic objectives. Gary www.hospital-robots.com Gary is a wheeled hospital robot from Israeli company Unlimited Robotics – see Fig.40. It can perform tasks such as providing bedside assistance to staff, medication reminders, provide companionship to patients, deliver supplies, sanitise, clean, automate mundane tasks etc. It runs on an Intel Core i7 processor with 512GB of memory and has a Google tensor processing unit (TPU) for its AI for neural network machine learning. For software, it uses the Linux operating system with the in-house developed Ra-Ya platform, which is designed to make it easier for inexperienced robot programmers to build applications if they are familiar with Python or JavaScript. Geminoid HI-6 In 2006, Professor Hiroshi Ishiguro of the Intelligent Robotics Laboratory Fig.38 (below): the HUBO2 robot. Source: www.rainbowrobotics.com/en_hubo2 at The University of Osaka made an android replica of himself, the Geminoid HI-1. It has now upgraded to the Geminoid HI-6 – see Fig.42. It is an upper body only and is teleoperated. It is pneumatically operated with 16 actuators and has 16 degrees of freedom. The purpose of the android is to explore questions of “what exactly human presence is, whether human presence can be transmitted to remote locations, and whether androids can surpass humans through experimentation”. GR-1 & GR-2 www.fftai.com Fourier is a Chinese company that has developed the GR-1 and GR-2 humanoid robots. The GR-2 is a general purpose robot with sensors in its dextrous hands to make them touch-sensitive. It can be programmed with frameworks like ROS and NVIDIA Isaac Lab. See the video at https://youtu.be/N7qYcOuR7P8 HMND 01 https://thehumanoid.ai Humanoid has developed the HMND 01 robot, which they describe as a “labour automation unit” (see Fig.39). It is 175cm tall, weighs 70kg, has 41 degrees of freedom, a payload capacity of 15kg, a runtime of four hours and walking speed of 5.4km/h. Uses for the robot include goods handling, such as in warehouses, picking and packing for e-­commerce warehouses and parts handling in manufacturing operations. HRP-5P An experimental humanoid robot developed in Japan in 2018 by the National Institute of Advanced Industrial Science and Technology, it was demonstrated installing wall plasterboard sheeting similar to Gyprock. See https:// youtu.be/fMwiZXxo9Qg HUBO www.rainbow-robotics.com Rainbow Robotics is a Korean company that offers the HUBO2, shown in Fig.38, Fig.39: the HMND 01 robot. Source: TechNode – siliconchip.au/link/ac7p siliconchip.com.au which they claim is the world’s first humanoid robot to be commercialised. The first HUBO was released in 2004, with the HUBO2 going on sale in 2010. It has made commercial appearances, such as at the 2012 Philadelphia Phillies baseball game. iCub https://icub.iit.it iCub is an open-source research and recreational robot designed by a consortium of European universities for research into human cognition and artificial intelligence – see Fig.41. It is 104cm tall and weighs 22kg. It has been demonstrated with capabilities such as crawling, solving 3D mazes, performing archery, producing facial expressions, exercising force control, grasping small objects, and performing collision avoidance. It is said to use a neuromorphic processor. Fig.40: Gary, the hospital robot, enters a room. Source: www.hospital-robots. com/about Fig.41: iCub at the Center for Robotics and Intelligent Systems (CRIS). Source: https://w.wiki/Eotd Iggy Rob www.igus.com A partially humanoid robot; instead of having legs, it has a wheeled base (see Fig.43). Its suggested applications include performing wait staff duties, such as delivering food and drinks to restaurant customers, parts delivery on factory floors, and for education and research into robotics. It costs about US$54,500 ($83,500), is 1.7m tall and can carry 100kg. InMoov https://inmoov.fr The InMoov humanoid robot project is for hobbyists and universities, with a whole community of developers, including in Australia and New Zealand – see Fig.44. It is open source and can be 3D printed on any standard printer with a 12 × 12 × 12cm area. It utilises two Arduino Mega or Arduino Uno microcontroller boards, two Nervo Board shields and 28 servo motors. It has two cameras for object and face tracking, speakers for speech, one Kinect sensor (discontinued) or OAK-D-Lite-FF for 3D depth and gesture recognition, and a PIR sensor for presence detection. All of its fingers are motorised. Iron Iron from car manufacturer Xpeng (siliconchip.au/link/ac7i) uses their proprietary 40-core Turing AI chip with 3000 TOPS (trillions of operations per second) of processing power and their Tianji AIOS AI operating system, which is also used in their cars. It features 60 joints, 200 degrees siliconchip.com.au Fig.42: Professor Hiroshi Ishiguro with the android robot replica he made of himself, Geminoid HI-6. Which one is which? Source: https://drive.google.com/ drive/folders/1RN710FOs7r9KJ2TmXkh-W_0cqcHUtlNj Fig.43: the Iggy Rob robot. Source: www.igus.com/automation/news/ humanoid-robot Australia's electronics magazine Fig.44: the torso of InMoov. Its lower legs have not yet been developed. Source: https://inmoov.fr/gallery-v2/ December 2025  23 Figs.45, 46 & 47: the Kuavo 4.0 robot (source: www.lejurobot.com/en); the NASA Valkyrie or R5 robot, which is being tested in Australia (source: https://x.com/CWeezyeth/status/1643599326650986496/photo/1); and a range of humanoid industrial robots from Persona AI; a miner, builder, welder, fabricator and assembler (source: https://personainc.ai). of freedom, is 173cm tall and weighs 70kg. It is already being used to produce Xpeng cars – see Fig.48. Kuavo 4.0 www.lejurobot.com/en Kuavo is a product of Leju Robotics, designed as a general-purpose humanoid robot for applications such as personal assistance and industrial automation (Fig.45). It utilises Huawei’s Harmony­OS and PanGu multimodal large language model for AI. It is not clear if Kuavo uses it, but other products of this company are the world’s first to use 5G-A positioning, which uses 5G technology for high-precision location within indoor spaces. NASA The 2011 NASA Robonaut2 or R2 was the first humanoid in space, tested aboard the International Space Station (ISS). It was initially only a torso, but a mobility platform was added in 2014. It had significant technical problems, was not used and returned to Earth in 2018. The NASA Valkyrie (see Fig.46), also known as R5, is still under active development. It is also being tested in Australia by Woodside Energy “to develop remote mobile dexterous manipulation capabilities for uncrewed and offshore energy facilities”. It is 1.8m tall, weighs 136kg and runs on three Intel Core i7 CPUs. It is not currently deployed in space. Nurabot www.foxconn.com/en-us Foxconn, in cooperation with NVIDIA, has developed a robotic nurse with Tawian’s Taichung Veterans General Hospital called Nurabot (Fig.49). It has perception, navigation, understands language and has an ability to adapt. It is intended to address labour shortages, monitor patients’ vital signs, address caregiver burnout, move patients, deliver meals and medication, offer companionship, turn patients over in bed and learn patients’ habits. It does this with high-­ resolution sensors, autonomous navigation capabilities and NLP (natural language processing). In trials, the nurses love it because it takes over repetitive tasks and gives them more time for non-routine tasks. Patients like it as well, as the robot is always there for them. Optimus www.tesla.com/en_eu/AI Tesla has developed the Optimus Gen 2 general purpose humanoid robot Fig.48: Xpeng’s Iron robot making cars. Source: https://baa. Fig.49: the Nurabot nursing robot. Source: www.honhai. yiche.com/xiaopengP7jia/thread-51117011.html com/en-us/press-center/press-releases/latest-news/1605 24 Silicon Chip Australia's electronics magazine siliconchip.com.au Fig.50: the Optimus robot through various design iterations. Left to right, they are Bumblebee (September 2022), Optimus Gen 1 (March 2023), Gen 2 (December 2023) and Gen 3 (current). Source: https://x.com/ niccruzpatane/status/ 1937798034894762071/ photo/1 shown in Fig.50. The status of Gen 3 shown in that figure is unclear. Optimus uses Tesla’s in-house AI and uses parts of Tesla cars, such as the same computers, with custom-­ designed chips to provide the robot’s neural networks for tasks like locomotion, manipulation, navigation, vision processing and decision-making. Tthe same AI architecture as Tesla’s Autopilot and FSD systems. It uses the same cameras as Tesla cars for vision processing, actuators based on design principles developed for the cars, and the same 4680 batteries used in the Cybertruck. The neural networks used by Optimus are the same as used by Tesla cars to process visual inputs and adapt to environments. It also uses the same deep learning and auto-labelling techniques used by vehicles. Using its neural network, it can learn from videos of humans performing various tasks such as vacuuming, stirring food, disposing of rubbish or moving items on a factory floor. Elon Musk has previously shared a vision of collective learning and data sharing in Tesla’s AI and robotics initiatives, so the possibility exists that when one Optimus robot learns a new skill, they could all learn it, depending on the programming. Some critics have pointed out that early demonstration videos of Optimus involved teleoperation by human siliconchip.com.au operators. Tesla is currently using two Optimus robots in its factories for tasks like moving batteries, but they currently operate at half the efficiency of human workers. They intend to produce 1000 robots by the end of 2025, to be used by Tesla, with sales to other companies starting in 2026. The estimated cost per unit is US$20,000 to US$30,000 ($30,000 to $45,000). There has been discussion of controlling an Optimus robot via patients with the Neuralink brain-computer interface implant. This means a disabled person could ‘inhabit’ the body of an Optimus robot and control it to perform tasks by thought alone. It has also been proposed that such a person could control just parts of the robot, such as the arms, legs or hands, which could be attached to the body as artificial limbs. A patient named Alex has already demonstrated the ability to control an Optimus hand through Neuralink, via thought alone; see the video at siliconchip.au/link/ac7n Optimus has been integrated with X’s Grok AI, so it will be possible to have conversations with the robot and to get the Optimus to do anything as commanded by the operator. It may also be able to provide companionship. PAL Robotics https://pal-robotics.com This Spanish firm makes a range of Australia's electronics magazine robots, including bipedal humanoid types such as the REEM-C, TALOS and Kangaroo research platforms for robotics and AI research. They also make the ARI wheeled social robot for tasks like answering customer enquiries or running promotional campaigns. It uses AI, running on Ubuntu and the ROS framework, with facial recognition and other visualisation tools. Persona AI https://personainc.ai Persona AI in Houston are developing a range of heavy-duty industrial humanoid robots, based on one modular platform but customised for a variety of tasks including shipbuilding and welding – see Fig.47. One of the strengths of their robots is their advanced hands, derived from work by NASA. The firm was co-founded by ex-NASA staff. One usage model for these robots includes renting them out for specific jobs. For deployment for precision shipyard welding, they have a development partnership with HD Hyundai Robotics, Vazil and HD Korea Shipbuilding & Offshore Engineering, and are expected to be deployed in 2027. December 2025  25 Phoenix www.sanctuary.ai A robot from Canadian company Sanctuary AI (Fig.51), according to the company, Phoenix, running under their Carbon AI control system, “mimics subsystems found in the human brain, such as memory, sight, sound, and touch”. See the video at https:// youtu.be/-HizP4UQvug for more information. Fig.51: the upper body of the Phoenix robot. Fig.54: the OP3 robot. Source: https://en.robotis.com/sub/ business_platform_ op3.php Fig.52: the EngineAI PM01 in service as an experimental police robot. Source: South China Morning Post – siliconchip.au/link/ac7q Fig.55: a robot hand from Shadow Robot. Source: https://shadowrobot. com/dexterous-hand-series Fig.56: SoftBank Robotics’ NAO. PM01 www.engineai.com.cn EngineAI makes the PM01 general-­ purpose humanoid robot. Chinese police are using one in an experimental and demonstration role (see Fig.52). In police work, it can perform facial recognition and crowd scanning, although at this stage of development, it is not likely to be intelligent enough to be genuinely useful for police work. Still, it hints at a possible Robo­Coplike future. See the video at https:// youtu.be/vu930qj9CEI Poppy Project www.poppy-project.org An open-source platform for the creation, use and sharing of interactive 3D-printed robots for education, artists, scientists and hackers (see Fig.53). Porton Man Robotic Test System The Porton Man Robotic Test System (siliconchip.au/link/ac7j) is a humanoid robot for the US Army. Its purpose is to wear and test nuclear, biological and chemical (NBC) protective suits. It performs tasks that soldiers would perform, such as walking, running, kneeling, or laying prone while wearing the equipment and has 100 embedded sensors to test for leaks and to measure other parameters. Its main advantage is the ability to repeat movements precisely, enabling effective comparisons between different NBC ensembles. RoboPrime humanoid project RoboPrime is a very low-cost 3D-printed humanoid robot project for the enthusiast. The website at https:// github.com/simonepri/roboprime is no longer actively maintained; however, builders might still find some ideas there. Fig.53: the Poppy Humanoid v1.0.2 is 83cm tall, weighs 3.5kg, has 25 actuators and runs Odroid XU4 with Ubuntu 14.04. Source: www.poppyproject.org/en/robots/poppy-humanoid Fig.57: SoftBank Robotics’ Pepper. 26 Australia's electronics magazine Silicon Chip ROBOTIS https://en.robotis.com This Korean robot company makes open-source humanoid platforms, such as the OP3 (Fig.54), for research siliconchip.com.au Humanoid robots losing control There were recent incidents of humanoid robots losing control, which can happen, just like any other machine. There is obvious potential for this to harm people. That is why it is imperative that these robots are built with failsafe systems and some means for their handlers to deactivate them. You can see one such incident in the video at https://youtu.be/1eYZr9vdGl8 Legal concerns for AI There are no specific laws governing AI in Australia. In the event that AI and humanoid robots become sufficiently advanced to develop consciousness, it has been argued that they should be afforded rights as humans have. However, they are still man-made machines that mimic humans and still not human, just very advanced appliances. In regards to foundation models, polls of organisations have suggested that there is agreement that those who develop the models should be responsible for the risks, not those who use the models. and education purposes. You can find the relevant files at siliconchip.au/ link/ac7k Shadow Robot Company The Shadow Robot company (https://shadowrobot.com) specialises in making dextrous robot hands for other robotics manufacturers. One of their robot hands is shown in Fig.55. SoftBank www.softbankrobotics.com SoftBank Robotics offer two humanoid robots, NAO and Pepper. NAO (Fig.56) is described a teaching assistant. It is described as having a personality and an ability to inspire students from all ages, from preschool to university, including the ability to work with special-needs children. It is described as being able to connect “theory to practice with hands-on Fig.58: Sophia by Hanson Robotics. Source: www.hansonrobotics.com/ sophia siliconchip.com.au projects that encourage participation, teamwork, and creative problem-­ solving”. NAO can speak 20 languages, move naturally and runs on the Linuxbased NAOqi OS, a flexible framework that gives a lot of options for customising the robot. It is quite small at around 57cm tall, with a weight of 4.8kg. Pepper (Fig.57) is a robot designed to greet customers in a business. It can make personalised recommendations, help people find what they’re looking for, sell to and interact with humans. Pepper is 1.2m tall, weighs 28kg, runs for 12 hours on one charge, has a variety of sensors, runs on the NAOqi OS and can be customised with various SDKs (software development kits). carry on human-like conversations. The manufacturer describes Sophia as “a human-crafted science fiction character depicting the future of AI and robotics, and a platform for advanced robotics and AI research”. The robot has made appearances on many popular TV shows. Sophia utilises symbolic AI, neural networks, expert systems, machine perception, conversational natural language processing, adaptive motor control and cognitive architecture systems, among others. The manufacturer says that the robot has demonstrated rudimentary levels of consciousness under certain conditions using the Tononi Phi system of measuring consciousness (siliconchip. au/link/ac7l). StUWArt A 140cm-tall experimental humanoid robot from the University of Western Australia (UWA; siliconchip.au/ link/ac7m) – see Fig.59. It uses a commercial robot platform, with the focus of UWA research being on the development by students of autonomous control software enabling it to move and walk. The basic platform appears to be a Unitree G1, described below. Sophia www.hansonrobotics.com A humanoid robot by Hong-Kongbased Hanson Robotics (Fig.58). It can Titan www.roboforce.ai RoboForce has developed the Titan industrial robot, which can lift 40kg and place it with 1mm accuracy, running for eight hours on one charge – see Fig.60. The robots are modular and can be equipped with different hands Fig.59: the StUWArt humanoid robot with software under development by UWA students. Fig.60: the Titan-T industrial robot by RoboForce. Source: www.roboforce. ai/product Australia's electronics magazine December 2025  27 and bases. It is optimised to perform the five so-called primitive actions of all human labour: pick, place, press, twist and connect. Fig.61: the Unitree G1 robot. Source: www.unitree.com/g1 Fig.62: the soccer-playing version of the Unitree G1, the G1-Comp. Source: www.unitree.com/robocup Toyota Research Institute Toyota sees its future with large numbers of its workers being humanoid robots. The Toyota Research Institute has partnered with Boston Dynamics to integrate its ‘large behaviour models’ with Boston’s Atlas robot. Toyota has trained their in-house AI large behaviour model to perform 500 tasks, to be integrated with Atlas. Toyota trains their large behaviour models using humans to demonstrate the required tasks, with joysticks to control robot movement, or robots are trained using videos. The AI model then synthesises these experienced operations into the relevant actions. Toyota envisions using the robots for tasks such as transporting parts, assembling components and conducting inspections. They may eventually replace between 5% and 15% of human workers. There is a video about Toyota’s large language models at https://youtu. be/DeLpnTgzJT4 Unitree G1 robot www.unitree.com Standing at 130cm tall, and weighing 35kg, the G1 has a swappable battery pack with a life of two hours, three-finger hands with optional tactile sensor arrays, 3D lidar and depth cameras, a microphone, speakers, 43 degrees of freedom, eight CPUs and other features – see Fig.61. Another version of this robot, the G1-Comp, is designed for playing in robot soccer competitions (Fig.62). The G1-Comp uses the YOLOv11 algorithm for real-time object detection and recognition, pose estimation and image classification using convolutional neural networks. The H1 is similar to the G1 but is taller (180cm). See the videos at https://youtu.be/ GzX1qOIO1bE (G1) and https://youtu. be/M0KrTumJBFc (G1-Comp). Fig.63: the Walker S Lite (left) and Walker S. Source: www.ubtrobot.com/en/ humanoid/products/WalkerS Walker S www.ubtrobot.com The Chinese UBTECH Walker S Industrial Humanoid Robot assists on production lines with inspections and installing small parts. It comes in two models: the Walker S (170cm tall, 65kg, 41 degrees of freedom, 2.5 hour battery) and the Walker S Lite (130cm Australia's electronics magazine siliconchip.com.au 28 Silicon Chip What is AIoT? AIoT or artificial intelligence of things is like the internet of things (IoT), where devices can communicate with each other. However, each device also possesses artificial intelligence. The combination of IoT and AI in AIoT makes for an extremely powerful network, where AI-powered devices can interact and communicate with each other locally, area-wide, country-wide or even worldwide. It could conceivably be dangerous in the future, if not managed appropriately, with limitations to stop AI getting out of control. Just imagine an army of robots being programmed with malicious intent… Glossary of Terms AI – Artificial Intelligence; machines simulating human intelligence, such as learning, reasoning and problem-solving ANN – Artificial Neural Network; computational models inspired by human brains, used in machine learning ASIC – Application-Specific Integrated Circuit; a custom-designed chip optimised for a specific function or task CNN – Convolutional Neural Network; deep tall, 63kg, 41 degrees of freedom, two hour battery) – see Fig.63. Features of the Walker S include all-terrain autonomous adaptation, robust self-balancing, multi-modal large model-based decision making, hand-eye coordination and whole body manipulation, U-SLAM (UBTECH simultaneous location and mapping), 3D point cloud semantic navigation, human and environment comprehensive perception and multimodal human-robot interaction. It runs on Robot Operating System, Linux ROSA 2.0, supports teleoperation and AIoT (artificial intelligence of things). For more information, see the videos at https://youtu.be/UCt7qPpTt-g and siliconchip.au/link/ac7o carpentry, 3D concrete printing and other tasks – see Fig.64. It is intended to address a shortage of skilled construction workers and to reduce human injuries. Zyrex https://ricrobotics.com This 6m-tall AI robot for construction sites from the Californian company RIC Robotics is yet to be released, but is designed to perform welding, Further viewing The smallest humanoid robot The world’s smallest humanoid robot is 57.7mm tall and was built by Tatsuhiko Mitsuya at the Nagoya Institute of Technology in Japan – see Fig.65. Upgrading humanoid robots AI is progressing rapidly. Since AI is software-based, it is usually possible to upgrade a humanoid robot to make it smarter as new AI software is released, thus protecting the investment in robot hardware. A fascinating look at “The Proto-­ Robots of Antiquity” is available in the YouTube video at https://youtu. SC be/0QGkf13fVs4 learning models optimised for vision, detecting edges, shapes and patterns CPU – Central Processing Unit; a general- purpose processor that executes instructions & manages computing tasks DoF – Degrees of Freedom; independent movements a robot joint or mechanism can perform End Effector – a tool/device at a robotic arm’s end that interacts with objects FPGA – Field-Programmable Gate Array; a chip programmable for specific hardware tasks post-manufacturing GPU – Graphics Processing Unit; a processor specialised for highly parallel tasks like machine learning LLM – Large Language Model; an AI model trained on massive text datasets to generate or understand language Multimodal – An AI that processes and integrates multiple data types (text, images, audio, video etc) Neuromorphic Processor – a chip that uses artificial neurons to mimic the human brain NLP – Natural Language Processing; an AI’s ability to understand, interpret and generate human language Organoid – a simplified version of an organ designed to imitate it RTOS – Real-Time Operating System; an operating system that guarantees timely processing for critical tasks Fig.64: the giant Zyrex construction robot, compared to a human-sized robot at lower left. The robot is not fully humanoid like the others in this article – but it has some similarities. Source: Robotics & Automation News – siliconchip.au/link/ac7r Fig.65: the world’s smallest humanoid robot. Tactel – Tactile Element; a sensor element that detects touch, pressure or texture information Teleoperation – operating a machine remotely TPU – Tensor Processing Unit; a Google- designed chip optimised for accelerating machine learning workloads. Transformer – a neural network architecture that uses attention to process sequential data efficiently VLA – Vision-Language Action; an AI that combines visual input and language to perform actions or tasks VLM – Vision-Language Model; an AI that combines image understanding with text comprehension and generation siliconchip.com.au Australia's electronics magazine December 2025  29