Foundation Tech Stack
General-purpose humanoid robots with reinforcement learning and custom silicon
Foundation is building general-purpose humanoid robots trained via in-house reinforcement learning pipelines, with a heavy tilt toward custom hardware and inference silicon. The tech stack spans embedded systems (ARM, RTOS, FreeRTOS, VxWorks), simulation-to-real training (MuJoCo, PyBullet, PyTorch), and robotics middleware (ROS 2), with active adoption of CUDA signaling GPU-accelerated inference work. The 64-engineer hiring mix (78 open roles, 52 posted last month) skews heavily toward embedded and ML infrastructure, matched by concurrent projects in RL training pipelines, custom SLAM, and inference chip evaluation—suggesting a company scaling from prototype toward production while building proprietary silicon to reduce cost and inference latency.
Tech Stack 59 technologies
What Foundation Is Building
◆Challenges
- Cost efficiencies
- Transition from prototype production to scalable manufacturing
- Off-the-shelf slam solutions insufficient
- Reducing human risk in conflict zones
- Eliminating recurring integration challenges
- Aggressive ramp targets
- Launching factory
- Evaluating custom inference chip feasibility
- Designing dexterous robotic hand
- Generalizing across unstructured settings
▲Active Projects
- Hardware-in-the-loop validation of learned policies
- In-house rl training pipelines and tooling
- Reinforcement learning algorithms for real-time control and locomotion of humanoid robots
- Custom electronics design for robotic systems
- Simulation-to-real training workflows
- Embedded systems and mixed-signal circuits design
- Custom embodied slams solutions
- Vision-language-action model development
- Factory launch
- Custom inference chip evaluation
Hiring Activity
Department
Seniority
Notable leadership hires: Chief Designer, Motor Actuator Lead
