Planned synchronization for multi-robot systems with active observations

Area of Science:

• Robotics and Artificial Intelligence
• Multi-Agent Systems
• Planning Under Uncertainty

Background:

• Cooperative multi-agent robotic systems require joint action planning based on shared state observations.
• Planning perception and communication acts under uncertainty is crucial for efficient cooperation.
• Existing methods struggle with the intractability of large joint belief spaces.

Purpose of the Study:

• To develop a computationally tractable approach for multi-agent planning under uncertainty.
• To enable robots to decide proactively when the cost of obtaining state information is justified.
• To formulate a method suitable for high-quality observations that recover joint states, even if infrequently.

Main Methods:

• Formulation of the problem as a Markov decision process (MDP) solved over macro-actions.
• Development of a Bellman-like recurrence to guide policy generation.
• Policies simultaneously define low-level actions, state recovery stages, and future rescheduling commitments.

Main Results:

• The proposed MDP formulation effectively sidesteps the need for full joint belief space construction.
• Demonstrated multi-agency in practical forms: assistance, sensor fusion, and coordinated activity.
• Successful simulation studies and physical robot implementation validated the approach for decentralized execution.

Conclusions:

• The developed method provides an effective framework for decentralized execution of joint plans in multi-agent robotic systems.
• The approach is adaptable to real-world non-idealities, as shown by enhancements based on hardware experience.
• This work advances cooperative robotic capabilities in scenarios requiring joint state estimation and action planning.