Moby

Moby is a pioneering lightweight humanoid robot developed by Under Control Robotics (UCR), a Silicon Valley startup founded in April 2024 by a team of robotics experts with backgrounds from NASA, Apple, Caltech, and SpaceX. The initial prototype was remarkably built from concept to functional walking robot in just four months by a team of four, marking a significant achievement in rapid prototyping for humanoid robotics. Designed primarily for hazardous industrial environments such as construction sites, mining operations, offshore oil rigs, and energy facilities, Moby addresses critical needs where human workers face high risks from toxic fumes, unstable terrain, falls, and costly downtime. Unlike high-end specialized robots, Moby emphasizes affordability, ruggedness, mass-producibility, and adaptability without relying on heavy supercomputers or exotic hardware. Architecturally, Moby features a bipedal humanoid form factor standing at 170 cm tall and weighing 45 kg, constructed from composite polymers and an aluminum frame for lightweight durability (IP20 ingress protection). It boasts 34 degrees of freedom overall, including 10 in each hand with 10 fingers for dexterous manipulation. Propulsion comes from brushless motors paired with harmonic or planetary gears, enabling a top walking speed of 3 km/h and runtime of up to 5 hours per charge. Early versions relied on low-cost smartphone-grade IMU sensors for balance and orientation, eschewing arms for initial bipedal locomotion tests. Subsequent iterations, including Moby V2 demonstrated publicly in March 2025, integrated dual arms capable of handling up to 60 lbs (27 kg) payload—far exceeding initial single-arm 3 kg / dual 9 kg specs—while maintaining stability across diverse terrains. Moby's AI architecture is a standout, powered by a unified lightweight AI model optimized for edge computing on NVIDIA Jetson Orin platforms with Arm Cortex-A78 CPUs and STM32 motor controllers. This model, trained extensively in simulation using NVIDIA's Isaac Sim, Isaac Lab, GR00T foundation models, and X-Mobility policy (a vision-based end-to-end imitation learning system with residual RL), achieves seamless sim-to-real transfer with 95% success rates. The pipeline includes real2sim for actuator dynamics, sim2sim cross-simulator consistency, sim2real domain randomization, and real2real fine-tuning with memory-driven adaptation. Vision integration in later versions enables real-time perception, obstacle avoidance, and adaptive navigation. The system supports ROS on Ubuntu Linux, Wi-Fi/Bluetooth connectivity, and LLM integration for enhanced decision-making. Demonstrations showcase Moby's robust mobility: omnidirectional walking, rough terrain handling (grass, debris), stair climbing, ramp ascent/descent (including backwards), side-stepping, and resilience to kicks or shoves without cameras in early tests. Manipulation capabilities focus on rugged tasks like material handling in harsh conditions. As of late 2025, Moby remains in prototype stage with V2 advancements, no full commercial deployments reported, but field trips and lab demos validate its potential for immediate industrial use—deployable without PPE or oxygen, reducing million-dollar-per-hour downtime. UCR's lean approach positions Moby as the most inexpensive full humanoid in America, targeting research, education, hospitality, and heavy industry markets with developer-friendly customization.