Phantom MK1

The Phantom MK1 is a pioneering camera-first humanoid robot developed by Foundation Robotics, a San Francisco-based startup founded in 2024 by CEO Sankaet Pathak, co-founder Mike LeBlanc (former US Marine), and others. Following the acquisition of Boardwalk Robotics in December 2024, Foundation accelerated development, building the Phantom MK1 from concept to first shipment in just 13 months, with the prototype publicly debuting in February 2025 at San Francisco's Temple Nightclub where it performed as a DJ with preprogrammed motions. By December 2025, over 40 units had been built and delivered, with initial customer pilots starting in April 2025 and scaling plans for 10,000 units in 2026. Architecturally, the Phantom MK1 stands 175 cm (5'9") tall and weighs approximately 80 kg (176 lbs), constructed from steel and plastic for durability in rugged environments. It features 19 degrees of freedom in the upper body, powered by advanced cycloidal actuators delivering up to 160 Newton-meters of peak torque while maintaining back-drivability below 1 Newton-meter, ensuring safe interaction with humans. The anthropomorphic hands have 10 fingers for precise manipulation. Mobility includes a top walking speed of 1.7 m/s (6.1 km/h), enabling navigation in human-built spaces without modifications. Perception relies on a camera-first approach with eight high-resolution cameras in the head for vision-based AI, supporting tasks like obstacle avoidance, grasping, and tool use. The AI stack is LLM-driven, combining large language models for high-level task reasoning with proprietary physics-action models that translate natural language instructions into precise whole-body motions. This task-to-motion architecture allows decomposition of complex missions into sequences, with low latency from perception to action. A key innovation is fleet coherence via a shared world graph, where multiple robots share real-time environmental data, task status, and optimizations to prevent duplication, reallocate resources dynamically, and enhance collective intelligence—akin to GPU cluster coordination. Training leverages real-world deployment data in a human-in-the-loop feedback loop, improving autonomy over time and reducing teleoperation needs. Real-world deployments began with industrial pilots in automotive manufacturing (e.g., Atlanta automaker handling bumpers), consumer goods, beverage, and glass sectors, operating 24/7 in unstructured factories to address high worker attrition. Approximately 10 units are currently deployed, generating data for AI refinement under a Robotics-as-a-Service (RaaS) model. Defense applications, enabled by $10M in government contracts inherited from Boardwalk, include logistics, aircraft maintenance, barrier breaching, casualty evacuation, and reconnaissance. Foundation is in advanced talks with US Army, Air Force, Marines, and DHS, embracing dual-use potential including weapon proficiency training (e.g., C-4 placement, gunfire) to counter adversaries like China's Unitree robots. CEO Pathak envisions deployments on battlefields, borders, and eventually Mars/Antarctica bases, with MK2 slated for 2026 featuring enhanced actuators and 30% payload increase. Priced at ~$150k in small volumes, scaling promises cost halving. Challenges include battery life (2-3 hours estimated), reliability (demo falls noted), and ethical concerns, but experts see near-term viability in hazardous tasks. Foundation's bold master plan prioritizes humanoids for labor shortages amid declining birthrates, funding extraterrestrial expansion.