Design and maintain high-fidelity MuJoCo-based simulation environments used for training and evaluating AI-driven robotic systems. You'll develop and refine physics models, task configurations, and agent behaviors to support cutting-edge research in embodied intelligence.
What You'll Do
- Build and iterate on simulation environments using MuJoCo, dm_control, or Gymnasium-Robotics to model realistic robotic tasks
- Implement and fine-tune reinforcement learning algorithms such as PPO, SAC, and TD3 to train robust policies
- Define observation and action spaces, along with reward functions, that promote generalizable and stable agent behavior
- Diagnose and resolve issues in physics simulations, including contact dynamics, actuator responses, and scene setup
- Evaluate trained agents for performance, consistency, and potential for real-world transfer
- Document simulation designs, training workflows, and experimental outcomes with clarity and precision
- Collaborate asynchronously with research engineers, incorporating feedback into model improvements through reward shaping and iterative refinement
- Stay current with developments in robot learning, simulation techniques, and embodied AI frameworks
Requirements
- Proven experience working directly with MuJoCo or compatible frameworks like dm_control or Gymnasium-Robotics
- Firm grasp of reinforcement learning theory and hands-on experience with training pipelines
- Proficiency in Python and modern ML libraries such as PyTorch or JAX
- Experience crafting reward functions for complex, multi-joint robotic tasks
- Understanding of robot kinematics, dynamics, and control principles
- Ability to read, write, and interpret MJCF or XML model definitions and their physical implications
- Self-motivated with strong attention to detail and the ability to work independently
- Excellent written communication skills—essential for documenting decisions and explaining technical trade-offs
- Must verify identity and confirm eligibility to work as an independent contractor in your country
Preferred Qualifications
- Experience with sim-to-real transfer methods such as domain randomization or system identification
- Familiarity with alternative physics simulators like Isaac Gym, PyBullet, Drake, or Genesis
- Background in multi-agent systems or hierarchical reinforcement learning
- Published work or open-source contributions in robotics, reinforcement learning, or embodied AI
- Experience integrating imitation learning, model-based RL, or world models into training pipelines
- Graduate-level education in robotics, machine learning, computer science, or a related field
Technical Stack
MuJoCo, dm_control, Gymnasium-Robotics, PPO, SAC, TD3, Python, PyTorch, JAX, MJCF, XML, Isaac Gym, PyBullet, Drake, Genesis
Benefits
- Fully remote work from approved locations
- Weekly payments via PayPal or Stripe
- Flexible project-based schedule with availability from 15 to over 40 hours per week
Work Mode
This is a global remote role open to contractors in specific countries. You must be able to commit to at least 15 hours per week, with variable weekly hours depending on project needs. Collaboration is fully asynchronous, emphasizing self-direction and written clarity.
Compensation
Hourly rate ranges from $30 to $70, depending on location and experience level. Most projects fall around $30/hour. Payment is issued weekly via PayPal or Stripe.
