Humanoids
Daily
Published on

Foundation Emerges With 'Phantom' Humanoid, Betting on Novel Actuators and Hybrid AI

Authors
An illustration of Foundation CEO Sankaet Pathak surrounded by humanoid robots
Sankaet Pathak—India-born coder who wrote his first programs at 12, researched computer-vision in astrophysics labs, founded fintech firm Synapse, and now leads Foundation’s push into humanoid robotics. Image credit: Illustration generated with OpenAI Sora and some prompt-tweaking

Foundation Emerges With 'Phantom' Humanoid, Betting on Novel Actuators and Hybrid AI

Foundation, a robotics company founded in May 2023 by former Synapse CEO Sankaet Pathak, is stepping into the increasingly competitive humanoid robotics arena. With offices in San Francisco and Munich, the company is developing 'Phantom', a humanoid robot designed primarily for industrial labor tasks in sectors like manufacturing, logistics, and potentially defense and space construction.

A Different Approach to Hardware

Foundation is taking a full-stack approach, developing both its hardware and AI software in-house. A key claimed differentiator lies in its proprietary actuator technology. Unlike the harmonic or cycloid drives common in other humanoids, Foundation utilizes rolling contact gearboxes. Pathak asserts these offer significantly higher energy efficiency (claiming 90-95% vs. 50-60% for competitors), greater torque density enabling a roughly 20kg (45lbs) payload, reduced heat and wear, and inherent compliance, potentially making the robots safer for operation near humans. This hardware choice, the company believes, also simplifies the AI training process by improving simulation accuracy (sim-to-real transfer).

Every new sensor adds noise—keep it simple.

Initially, Phantom robots deployed in industrial settings will be tethered for power, though battery-powered versions are planned for the future. For perception, Foundation is following Tesla's lead, relying solely on cameras and eschewing LiDAR or other sensors to reduce system complexity.

Hybrid AI Strategy

On the software side, Foundation distinguishes itself with a hybrid AI strategy. Instead of relying solely on imitation learning (learning by demonstration), favored by competitors like Figure AI and Tesla's Optimus project, Foundation combines imitation learning with state-based models. These models explicitly incorporate knowledge of physics, kinematics, and task dynamics. Pathak argues this hybrid approach allows for more robust autonomy with significantly less training data – claiming simple tasks can be learned in about 30 minutes– reducing the need for large-scale teleoperation fleets during rollout.

In 12 months, humanoid robots might become the new MVP hardware form.

For locomotion, the company employs reinforcement learning (RL), which Pathak believes is rapidly maturing and could soon make bipedal robots easier and potentially cheaper to build than traditional wheeled autonomous mobile robots (AMRs).

Execution Focus and Ambitious Goals

Leveraging talent acquired from robotics research groups associated with IHMC in the US and the Technical University of Munich (including Prof. Patrick van der Smagt's team), Foundation aims for rapid iteration and deployment. Pathak, drawing lessons from his time scaling the FinTech company Synapse, emphasizes a culture of intense focus, execution velocity, and high talent density, openly stating that early-stage startups are incompatible with traditional work-life balance.

Your only real moat is execution velocity.

Having raised approximately $12 million in seed funding pre-product, Foundation is targeting initial customer deliveries of its Phantom robot around April or May 2025. Early use cases focus on automotive manufacturing (like racking/unracking parts) and logistics (moving boxes). The company aims to achieve a full shift of autonomous work in an automotive setting shortly after initial deployment. Scaling plans are ambitious: shipping 40-50 robots in 2025, with a target of over 10,000 units in 2026, contingent on execution. The target price is around $100,000 per robot via a subscription model, though initial costs might be closer to $150,000.

Foundation Phantom humanoid robot working in a warehouse
Phantom mk1 getting some work done. Image credit: Foundation

Watch Robert Scoble's tour of the Foundation HQ and interview with Sankaet Pathak here.

Navigating the Competitive Landscape

Foundation enters a field populated by increasingly high-profile players. Pathak identifies Figure AI and Tesla as primary competitors focused on deploying industrial humanoids with urgency. While acknowledging the scale of Chinese robotics manufacturing, Foundation believes its technological approach – particularly its actuators and hybrid AI – provides a competitive edge in efficiency, safety, and speed-to-autonomy.

The long-term vision includes robots capable of understanding and executing complex tasks via natural language voice commands, moving faster as AI improves, and potentially achieving zero-shot learning capabilities. Pathak also envisions augmented reality interfaces, like Apple's Vision Pro, playing a role in human-robot collaboration and teleoperation.

Foundation's strategy hinges on its ability to execute on its distinct hardware and AI choices and rapidly scale production. Its success will depend on proving its technological claims and deployment speed in real-world industrial environments.

Watch an interview with Sankaet Pathak here:

Sources:

Discuss on X
Subscribe to the newsletter