Researcher, Manipulation

Menlo Research is an Applied R&D lab building Asimov, an open-source humanoid robot platform, and the full software stack that powers it. Our mission is to make humanoid labor economically viable, turning software into physical labor at scale. We build across the full stack: hardware architecture, locomotion, autonomy, simulation, and infrastructure. We move fast, ship to real robots, and open-source everything we can. If you want your work to matter beyond a paper or a demo, this is the place.

We are looking for a Researcher, Manipulation to give Asimov hands that can do real work. We recently ported our RL stack from locomotion to manipulation, taking Asimov from walking to flipping, and you will drive what comes next: grasping, dexterous and contact-rich tasks, and the kind of whole-body coordination where an arm, a torso, and two legs have to solve one problem together. You will get skills onto real hardware and keep pushing until they hold up outside a controlled setup.

• Design, train, and ship reinforcement learning policies for manipulation, grasping, and dexterous, contact-rich tasks on Asimov
• Build on our locomotion-to-manipulation RL work and extend it toward whole-body manipulation, where balance and reaching are one problem
• Own the sim2real pipeline for manipulation, using our zero-shot sim2real approach so skills transfer to hardware cleanly
• Design rewards, training environments, and randomization strategies in MuJoCo for contact-heavy tasks that are hard to simulate well
• Close the loop with real telemetry from real arms and hands, then feed failures back into the next policy
• Collaborate closely with locomotion, controls, and hardware researchers, since a humanoid manipulates with its whole body
• Open-source what you can and share what you learn with the community building on Asimov
• Deep hands-on experience with reinforcement learning or learning-based methods for manipulation, grasping, or dexterous control
• A track record of getting manipulation policies onto real robots, not only into papers or simulators
• Strong command of a physics simulator such as MuJoCo, Isaac, or similar, including contact modeling, reward shaping, and domain randomization
• Fluency in Python and modern RL tooling, and comfort in a ROS2-based control stack
• A bias for shipping: you would rather watch a policy fumble a real object this week than perfect it in sim next month
• Clear thinking about why contact-rich tasks fail, and how to close the gap
• Published work in manipulation, dexterous control, grasping, or imitation learning
• Experience with tactile sensing, force control, or teleoperation and demonstration data
• Experience with whole-body control that spans locomotion and manipulation
• Contributions to open-source robotics or RL projects

…