Controls & Reinforcement Learning Engineer

Foundation•San Francisco, CA

11h•$150,000 - $280,000

About The Position

We're building a high-degree-of-freedom tendon-driven robotic hand, and we're looking for the first software engineer dedicated to it. This is a rare opportunity to own the entire controls and learning stack for one of the hardest manipulation problems in robotics — from low-level motor control to grasp planning — with direct impact on the product from day one. You'll work with a small team and have significant autonomy in how you approach problems. We're not looking for someone who needs a roadmap handed to them; we're looking for someone who can build one.

Requirements

Strong foundation in both classical control theory and modern RL — you're comfortable reasoning about stability and dynamics as well as policy optimization
Hands-on experience deploying learned policies on physical hardware, not just in simulation
Proficiency in Python and C++; familiarity with ROS/ROS2 and real-time systems
Experience with physics simulators (MuJoCo, Isaac Lab, or similar) and deep learning frameworks (PyTorch, JAX)
High agency and comfort operating independently in an early-stage environment — you define the problem as much as you solve it

Nice To Haves

Prior experience with tendon-driven or cable-actuated systems
Experience with dexterous manipulation, multi-fingered hands, or compliant mechanism control
Familiarity with tactile sensing integration
Contributions to open-source robotics or RL projects
M.Sc. or Ph.D. in Robotics, Controls, Computer Science, or a related field

Responsibilities

Low-level motor control for a tendon-driven, high-DOF hand system, including tension management, coupled joint dynamics, and real-time feedback loops
Higher-level coordination across actuators to achieve stable, dexterous finger and wrist trajectories
Integration of tactile, proprioceptive, and other sensor modalities into the control architecture
End-to-end RL pipeline for grasp planning and manipulation — from simulation training through sim2real transfer and physical deployment
Tooling, testing infrastructure, and software architecture decisions for the hand software stack