Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper presents a reinforcement learning framework for robotic strawberry harvesting that addresses the challenges posed by occlusions and deformable plant structures through a sequential decision-making approach. By integrating a shared interaction-aware policy with heuristic phase coordination, the system effectively manages obstacle separation, fruit detachment, and placement, achieving success rates of 89.7% in simulation and 82.0% in real-world tests. The methodology also supports zero-shot sim-to-real transfer, showcasing the adaptability of the learned behaviors across different robotic platforms despite increasing complexity due to occlusions.
Robotic strawberry harvesting can achieve over 82% success in real-world conditions, even in the face of severe occlusions and complex plant dynamics.
Severe occlusions and deformable plant structures introduce complex contact dynamics that challenge robotic strawberry harvesting. A policy-driven reinforcement learning (RL) framework with heuristic phase coordination was developed, in which obstacle separation, fruit detachment, and placement were formulated as a sequential decision-making task. A shared interaction-aware policy generated Cartesian motions across all task phases, while lightweight heuristic logic coordinated task progression and gripper events. A shared structured observation space was used to represent target, obstacle, end-effector, and task-context information. A hierarchical architecture combined the high-level policy with low-level Cartesian impedance control for compliant interaction. To support zero-shot sim-to-real transfer, feasibility-first observation alignment and domain randomization were adopted. The policy achieved success rates of 89.7% in simulation and 82.0% in real-world experiments. As the occlusion level increased from 1 to 5, the average execution time increased from 12.99 s to 21.73 s, reflecting greater interaction complexity. These results demonstrated effective transfer of interaction-aware harvesting behaviors to a structurally different robotic platform.