New robotic method for harvesting hydroponic lettuce
A new framework called Path-Safe Approach Proximal Policy Optimization reaches targets in constrained workspaces.
Researchers developed Path-Safe Approach Proximal Policy Optimization, or PSA-PPO. This framework is for robotic harvesting of nutrient film technique (NFT)-hydroponic lettuce in greenhouses.
The system was trained in Isaac Lab. It transferred to a physical UR10e robotic arm without real-world policy fine-tuning.
PSA-PPO uses target-conditioned observations and corridor-aware spatial constraints. It also uses trajectory-history-based safety encoding and task-specific success criteria.
In a physical benchmark with 150 scheduled targets, PSA-PPO achieved a 94.7% reaching rate. This outperformed Cartesian+TOTG at 86.0%, RRTConnect at 88.0%, and RRT* at 80.0%.
The method reduced median RMS jerk by 39.8 to 73.0% compared to classical planners. An integrated system with an RGB-D vision module achieved 94.4% vision-guided target reaching and 86.1% complete harvesting success.
Key facts
- PSA-PPO reached 94.7% of 150 scheduled targets.
- The system achieved 86.1% complete harvesting success in a complete greenhouse harvesting system.
- PSA-PPO reduced median RMS jerk by 39.8 to 73.0% relative to classical planners.