AI's next challenge is moving intelligence from digital environments into physical machines that can sense, decide, and act in the real world. Success in physical AI requires advancements across sensing, simulation, computing, connectivity, and control, not just better model performance. Physical AI systems must handle imperfect measurements, timing constraints, and the consequences of action. These systems move beyond analysis to predicting and producing useful actions amidst uncertain conditions. Analog Devices Inc. (ADI) is involved in this race, as its technologies bridge the physical and digital worlds.
Humanoid robots powered by AI are emerging as helpers rather than enemies, driven by advances in machine learning and perception. Robots like Boston Dynamics' Atlas learn through human demonstrations and simulations to master complex tasks, despite currently lacking full human-like abilities for daily life. The development is fueled by powerful AI, such as NVIDIA chips, which allow robots to process data and grasp the physical world. While there are concerns about job displacement and control, the focus is on training these systems with value systems and constraints. Ultimately, the goal is to create intelligent systems that augment human capabilities, not replace them entirely.
Lukas M. Ziegler highlighted Dexmate's Vega 1 Pro as a valuable robot research platform. This bimanual mobile robot costs $75,000 and features 24 degrees of freedom and a 5 kg payload per arm. Ziegler argues the Vega addresses a research bottleneck by providing ready hardware for policy testing. The platform includes Jetson AGX Thor compute, LiDAR sensors, and various modeling formats. This hardware is cited in major research initiatives like DeepMind and Amazon FAR.
RoboParty Robotics has released the RP1 humanoid as an open-source platform for robotics research and embodied AI development. The platform's main feature is providing access to its hardware, actuators, and control systems. This access allows researchers to adapt the system for experiments and test different control approaches. The focus is on providing a customizable system rather than a specific deployment or task demonstration. The intended applications are robotics experimentation and embodied AI development.
Agility Robotics, Moonlake, and Agility Robotics discussed practical constraints on robot deployment, focusing on simulation accuracy, onboard computing limits, and customer data rights. Simulation should serve to narrow testing rather than replace physical hardware tests, with challenges arising from modeling complex interactions like object contact. Economic and computational limits necessitate separating high-level planning from low-latency edge control. Furthermore, customer concerns regarding proprietary data and the need for reliable performance metrics were identified as significant constraints. Ultimately, better collection and reuse of deployment failure data are called for to improve subsequent model updates.
NeodraDynamics claims its 4-meter-tall, 1,200 kg Neo1 humanoid has a combined arm payload of 500 kg using electric motors and digital hydraulics. The robot is designed as a stationary industrial platform, not a walking humanoid. It features 41 independently controlled movements powered by miniature hydraulic actuators, each weighing 1 kg and capable of pushing loads over a metric ton. The company is currently testing the Neo1 on heavy factory tasks with JD Industrial. Intended applications for the robot include heavy component assembly, cargo handling, and disaster response.
1X Technologies' Neo humanoid was featured on TIME's Best Inventions of 2026 list for its role as a household helper. The $20,000 robot is designed to perform domestic tasks such as fetching objects and unloading a dishwasher. It weighs 66 pounds, can lift over 150 pounds, and can carry 55 pounds. Launched globally in 2026, it is positioned as a consumer product for home use rather than an industrial system. The robot is designed to autonomously handle various household chores and is capable of learning new tasks over time.