Sim-to-Real Reinforcement Learning for Path-Safe Target Reaching in Robotic Harvesting of Hydroponic Lettuce
The study introduced Path-Safe Approach Proximal Policy Optimization (PSA-PPO) for robotic harvesting of hydroponic lettuce, focusing on precise manipulation in constrained spaces. The method integrates target conditioning and safety constraints into the policy, proving that path safety and corridor shaping are crucial for successful harvesting. It achieved high reaching rates on physical robots while significantly reducing jerky movements compared to traditional planners.