Zero-Shot Sim-to-Real Reinforcement Learning for Fruit Harvesting

This paper presents a novel and comprehensive sim-to-real pipeline for autonomous strawberry harvesting from dense clusters using a Franka Panda robot. Our approach addresses the challenges of robotic manipulation in unstructured agricultural environments, particularly the difficulty of picking occluded and clustered fruit. We introduce "FruitGym", a custom open-source Mujoco simulation environment designed to train a deep reinforcement learning agent. This environment leverages extensive domain randomization, varying lighting, object placement, and sensor noise to ensure the learned policy is robust and can be transferred directly to the real world. The agent is trained using the Dormant Ratio Minimization algorithm, which enhances sample efficiency and exploration. The proposed pipeline bridges low-level control with high-level perception and decision making, demonstrating promising performance in both simulation and in a real laboratory environment, laying the groundwork for successful transfer to real-world autonomous fruit harvesting.