Project

The Problem: Traditional bipedal control is manually intensive, while physical hardware testing is slow, expensive, and prone to damage.

The Solution: This project utilizes a Sim-to-Real Reinforcement Learning pipeline to:

• Accelerate Development: Rapidly train walking policies using NVIDIA's IsaacLab simulator.
• Reduce Costs: Iterate on locomotion logic in a risk-free virtual environment before hardware deployment.
• Optimize Performance: Generate robust control policies that adapt to physical hardware constraints.

System

Conclusion

Lesson Learned

• Train a walking policy using Reinforcement Learning in IsaacLab
• Model and build a robot from ground up
• Interface electrical components using I²C

Future Work

• Connecting the Depth Camera to the walking policy so that the robot learns to avoid obstacles