Physical AI moves models off the data-center floor and into machines that see, move and touch. The hardware choices decide what the AI can do: a policy cannot outperform its cameras, its actuators' torque and backdrivability, its compute's latency, or its battery. This course gives engineers a working map of that hardware, organized around the questions a robot program has to answer.
You start with robot architectures, from arms and mobile manipulators to legged and humanoid platforms, and how degrees of freedom, payload and reach translate into a hardware bill of materials. You then compare onboard compute (NVIDIA Jetson AGX Thor and Orin, Qualcomm Dragonwing robotics processors, and offload to edge servers), focusing on what perception and vision-language-action workloads need rather than restating the Jetson internals covered in hw-206. The sensing module covers RGB, stereo and time-of-flight depth, LiDAR, IMUs, force-torque and tactile sensors, and synchronization; the actuation module covers motors, harmonic and planetary gearing, quasi-direct-drive and series-elastic actuators, and end effectors. The last modules cover real-time communication (EtherCAT, CAN FD, Time-Sensitive Networking, ROS 2 middleware), battery and thermal design, and functional safety under the 2025 editions of ISO 10218 and related standards. Product facts are current as of October 2026.