Reinforcement Learning Engineer, Grasping

Sangha Partners; Houston (TX); On-site; USD 120,000 - 170,000; Full time

Job summary

Sangha Partners seeks a Reinforcement Learning Engineer to advance dexterous grasping on real hardware with high-DOF robotic hands. You will study recent RL advances, design experiments, and implement sim-to-real pipelines for robust manipulation in MuJoCo and Isaac Sim.

The role emphasizes hands-on policy training, reward shaping, and evaluation on physical robots, with collaboration across software teams to deploy end-to-end grasping systems.

Qualifications

• BS/MS/PhD in Robotics, CS, ML or related field.
• 2+ years of RL for grasping with reward design, exploration, and policy training.
• Experience deploying RL policies on real robotic hands.
• Ability to read, understand, and implement recent robotics/ML research.
• Experience with sim-to-real transfer: domain randomization or physics tuning.
• Proficiency in Python and DL frameworks (PyTorch, JAX) and RL libraries (rsl_rl, skrl).
• Experience preparing meshes and collision geometries for RL environments (MuJoCo/Isaac Sim).

Responsibilities

• Train and iterate RL policies for grasping tasks including functional grasping and in-hand manipulation.
• Implement sim-to-real transfer pipelines bridging simulation and hardware.
• Design rewards, exploration strategies, and curricula for grasping environments.
• Run experiments on real robots and in simulation to evaluate policy behavior.
• Stay current with research in learning-based grasping and adapt ideas to our platform.
• Collaborate with software team to deploy end-to-end grasping systems.
• Benchmark policies across object diversity and real-world uncertainty.
• Integrate tactile sensing and feedback into learned grasp policies.

Skills

Reinforcement learning

Python

PyTorch

JAX

MuJoCo

Isaac Sim

Domain randomization

Sim-to-real

Experimentation

Education

MS in Robotics/CS/ML

PhD preferred

BS accepted

Tools

MuJoCo

Isaac Sim

Python

rsl_rl

skrl

Job description

We are looking for a Reinforcement Learning Engineer to join our client's Manipulation team, focused on dexterous grasping. Their goal is to ship capable, reliable grasping policies on real hardware with high-DOF robotic hands. We are looking for someone who can follow recent advances in reinforcement learning and related learning-based methods, judge what is practically useful, and adapt those ideas on their platform.

Your Role:

• Train and iterate on reinforcement learning policies for complex grasping tasks including functional grasping, tool use, in-hand manipulation, and environment interaction.
• Implement and refine sim-to-real transfer pipelines to bridge the gap between simulation and physical robotic hand performance.
• Design reward functions and exploration strategies specific to grasp acquisition, along with curriculum strategies and training environments in MuJoCo and Isaac Lab.
• Run experiments on real robots alongside simulation, evaluating and debugging policy behavior on hardware.
• Monitor, evaluate, and adapt state-of-the-art research in learning-based grasping to deploy on our humanoid platform.
• Collaborate with the rest of the software team to deploy end-to-end grasping systems.
• Benchmark and evaluate grasp policies across object diversity, clutter scenes, and real-world uncertainties.
• Integrate tactile sensing and feedback into grasp policies for robust, force-aware manipulation.

We're Looking For:

• BS, MS, or PhD in Robotics, Computer Science, Machine Learning, or a related field.
• 2+ years of hands-on experience in reinforcement learning specifically applied to grasping- reward design, exploration strategy, and policy training built around picking up and manipulating objects (not general manipulation, locomotion, or whole-body control)
• Experience deploying RL-trained policies on physical robotic hands- real hardware validation, not simulation-only work.
• Demonstrated ability to read, understand, and implement ideas from recent robotics and machine learning research.
• Experience with sim-to-real transfer: domain randomization, physics tuning, or real-world policy validation on hardware.
• Proficiency in Python and deep learning frameworks (PyTorch, JAX), along with RL libraries such as rsl_rl or skrl.
• Experience preparing meshes and collision geometries for RL environments in simulators such as MuJoCo and/or Isaac Sim.

Bonus Qualifications:

• Experience with tactile sensors and integrating tactile feedback into learned grasp policies.
• Experience with contact-rich manipulation and force/torque estimation.
• Familiarity with other learning-based approaches such as behavior cloning, imitation learning, or diffusion-based policy methods.
• Publications or project work at top-tier venues (CoRL, RSS, ICRA) on grasping or dexterous manipulation.
• Experience in a humanoid robot startup environment.