About Menlo
Menlo Research is an Applied R&D lab building the software and hardware stack for the humanoid century. Our products span Asimov, an open-source humanoid robot platform, and Menlo OS, an integrated suite for embodied AI development. Cyclotron is our sim-to-real locomotion training pipeline -- it trains locomotion policies in simulation and transfers them to physical humanoid robots.
The Role
The core thesis behind Cyclotron: sim-to-real is a data interface problem, not a fidelity problem. The Cyclotron Team builds the simulation infrastructure and learning systems that make that thesis hold up in production. As a Locomotion Engineer, you will own the policies and pipelines that drive bipedal locomotion on Asimov, from training in sim through deployment on hardware. We are looking for someone who has taken locomotion from algorithm to running robot -- someone who understands the hardware deeply enough to know what to optimize for and why. This is a lead-level role with end-to-end ownership across the stack.
What You'll Do
- Design and train RL locomotion policies for bipedal walking, lateral movement, push recovery, and dynamic motion
- Develop and maintain MuJoCo-based simulation environments with custom actuator models that faithfully reproduce hardware IO timing, CAN bus delays, and motor saturation characteristics
- Own the domain randomization strategy: model the specific sources of sim-real mismatch and randomize what is genuinely uncertain, not everything
- Run processor-in-the-loop validation, including real firmware against simulated robots over virtual CAN, before any policy touches hardware
- Deploy and tune locomotion controllers on Asimov, iterating rapidly across sim and physical platforms
- Work with Data Engine telemetry to refine domain randomization and close identified sim-real gaps
- Collaborate with firmware, mechanical, and perception teams on cross-stack issues affecting locomotion performance
- Contribute to open-source releases of locomotion models, training code, and simulation assets
What We're Looking For
- 5+ years of experience in locomotion, legged robotics, or related controls and learning fields
- Proven track record deploying locomotion on physical humanoid or legged robots in a production or near-production setting -- you have shipped policies that run on real hardware, not just in simulation
- Strong proficiency in Python and C++
- Deep understanding of reinforcement learning and sim-to-real transfer for humanoid locomotion, including policy architecture choices and training dynamics that hold up on real hardware
- Experience with trajectory optimization, MPC, or hybrid control approaches alongside learned policies
- Deep understanding of rigid body dynamics, contact mechanics, and whole-body control
- Hands-on experience debugging sim-to-real gaps on physical platforms -- you know what breaks and why
- Familiarity with actuator behavior, motor dynamics, and low-level control integration at the hardware level
- Familiarity with state estimation, sensor fusion, and motion capture for locomotion
- Strong experimental intuition and ability to drive complex cross-functional work from research through deployment
Bonus points for:
- Research background in legged robotics, RL for control, or related areas -- publications or preprints welcome
- Contributions to open-source robotics or locomotion research
- Experience with processor-in-the-loop or hardware-in-the-loop validation workflows
Why Join Menlo?
Cyclotron is where the physics get real. You will be working on one of the hardest open problems in robotics -- teaching a humanoid to move with the robustness and fluidity that real-world deployment demands. The team is small and the ownership is genuine: you will shape the simulation infrastructure, the training pipeline, and the policies that run on hardware. This is a role for someone who has been through the full cycle before and wants to do it again on an open platform where the work ships.
.jpg)
