Automotive Research
header foto

Generic Logistic Planning for Cooperative Agrobots using Master-Slave reinforcement learning

In the agriculture sector, using small, autonomous and cooperative machines, namely multi-agent agrobots, to take over specific tasks from farmers is an inevitable trend. To have optimal and reliable logistic planning for cooperative multi-agent agrobots, researchers from Farm Technology Group in Wageningen University propose a learning-based approach using Master-Slave Reinforcement Learning, that will be researched in the light of DurableCASE.

The main challenges of using multi-agent robots for harvesting are how to plan, communicate and cooperate multiple robots to solve a range of tasks efficiently and safely. Instead of using the traditional heuristic methods to solve a travelling salesman problem (TSP), the proposed approach aims to explore power of reinforcement learning in planning and coordinating generically, using a master learner (harvester) and the joint slave learners (trucks or chaser bins).

The KPIs will focus on number of required slave robots, master robot utilization, path coverage, operation time and energy usage. Soil compaction is also an important KPI.