Belief Consolidation in a Partially Observable Multi-Robot Environment with Limited Data Sharing Capabilities
Multi-robot coordination is crucial for reliable autonomy even in a limited-communication environment in which robots can communicate infrequently and can share limited amount of information between themselves. We develop a technique in which a robot-robot communication is triggered only when a significant event'' is encountered by a robot. A robot seeks information related to a particular event and the helping robot shares its knowledge about the same. Thus, a robot consolidates its belief about an event via gathering information from other robots in the environment. Depending on limited and erroneous sensing, a robot has partial observability of the state of the environment. We model this as a decentralized Partially Observable Markov Decision Process (POMDP) from the perspective of each robot which maintains a belief'' over the state of the environment. Belief is a probability distribution over the states of the environment. New observations, obtained by a robot itself or shared by other robots, evolve the belief of a robot. Considering the erroneous sensing capabilities, a robot needs to consolidate its belief about some significant state of the environment, at times. This enables identify a significant state with more certainty. Previous research proposes sharing all the unshared observations between a given pair of robots. However, limited connectivity restricts the robots to share all the unshared data. Our approach enables a robot consolidate its belief about the significant state by gathering information from its neighboring robots. While the likelihood of occurring the significant state is less than a pre-decided threshold, the robot requests its neighbors for selective information. Experimental results validate the efficacy of our approach for belief consolidation and thus identify significant events with higher likelihood.
This work has been done by me under the supervision of Dr. Tanmoy Kundu as part of B-tech project in 2024-2025