Robotics Simulation with Active Camera
Budget: £10 – £100 GBP
I'm developing a MuJoCo-based robot simulation in Python with an active camera (pan/tilt). The control loop is a BC–RL hybrid:
- Behavior Cloning (BC) for the robot arm actions
- Proximal Policy Optimization (PPO) for training the camera's pan/tilt policy to maximize task observability and performance
My requirements are to:
- Extend an existing baseline codebase
- Wire up the BC/RL loop
- Implement and integrate the camera actuator and observations
- Deliver a clean training pipeline with reproducible results
Nice to Have:
- Experience with Franka/Panda or similar arms in MuJoCo
- Prior work on active vision/gaze control or vision-based RL
- Comfort with MuJoCo 2.x/3.x, Gymnasium, PyTorch, and mujoco-py/mujoco Python APIs
- Behavior Cloning (BC) for the robot arm actions
- Proximal Policy Optimization (PPO) for training the camera's pan/tilt policy to maximize task observability and performance
My requirements are to:
- Extend an existing baseline codebase
- Wire up the BC/RL loop
- Implement and integrate the camera actuator and observations
- Deliver a clean training pipeline with reproducible results
Nice to Have:
- Experience with Franka/Panda or similar arms in MuJoCo
- Prior work on active vision/gaze control or vision-based RL
- Comfort with MuJoCo 2.x/3.x, Gymnasium, PyTorch, and mujoco-py/mujoco Python APIs