Endowing Runtime Self-adaptation to an Autonomous Pallet Truck

The need for greater flexibility, as well as new developments in navigation and cognition, have motivated the introduction of autonomous mobile robots (AMRs) in retail intralogistics and also in retail outlets. The autonomy of AMRs means that they can make continuous decisions on how to manage the mission taking into account the dynamics of a scenario where people and robots will share the same environment. To successfully accomplish this task, the AMR must autonomously monitor its own state, detect potential problems and react appropriately, modifying the course of action in real time when necessary. In this paper we describe our approach to provide an autonomous low-lift pallet AMR with run-time adaptability to internal and external context changes. The paper proposes the use of a two-steps fuzzy inference engine. The first one encodes a set of relevant non-functional properties using QoS metrics. The second one uses these metrics, and fuzzified context items, for evaluating the alternatives to the nominal course of action. Decision making is encoded into Behaviour Trees (BTs). The approach has been successfully endowed in a low-lift pallet robot from MetraLabs GmbH and tested in real-world environments.