Goal: Enable a swarm of 5 drones to navigate a forest environment without collision.
Problem: GPS signal is unreliable in dense canopies; traditional pathfinding is too slow.
Approach: Implemented a decentralized flocking algorithm combined with Lidar-based SLAM.
Result: Achieved 98% success rate in simulation environments.
Goal: Teach a bipedal robot to walk over uneven terrain.
Approach: Used PPO (Proximal Policy Optimization) with a custom reward function focusing on energy efficiency.
Result: Robot learned to stabilize within 2 million timesteps.