Systems and methods for probabilistic consensus on feature distribution for multi-robot systems with markovian exploration dynamics

A method, comprising:

• accessing, at a processor of a first robot in communication with a memory, sensor information from a sensor of a first robot, including a first time-step index and a grid-node identifier that denotes a first grid-node of a spatial grid indexed by grid-node identifiers, the first grid-node corresponding to a location of the first robot during capture of the sensor information;
• updating an occupancy vector over the spatial grid by setting an entry indexed by the grid-node identifier to an occupancy value indicative of a local detection outcome at the first grid-node;
• updating generating, for the first time-step index, first feature opinion information by normalizing the occupancy vector to obtain a first occupancy distribution over the grid-node identifiers;
• accessing second feature opinion information from a second robot over a local inter-robot network interface when the first robot and the second robot are co-located at the first grid-node, the second feature opinion information comprising a second occupancy distribution over the grid-node identifiers;
• generating a fused occupancy distribution by weighted element-wise combination of the first occupancy distribution and the second occupancy distribution;
• and
• updating the first feature opinion information for a subsequent time step based on the fused occupancy distribution.

A system, comprising:

• a plurality of robots, each robot of the plurality of robots being operable for motion within an area according to a Markovian random walk, the plurality of robots including a first robot, the first robot having a processor in communication with a memory, a sensor, and a local inter-robot network interface, the memory including instructions executable by the processor to:
• access sensor information from the sensor of the first robot, the sensor information including a time-step index and a grid-node identifier that denotes a first grid-node of a spatial grid indexed by grid-node identifiers, the first grid-node corresponding to a location of the first robot during capture of the sensor information;
• update an occupancy vector over the spatial grid by setting an entry indexed by the grid-node identifier to an occupancy value indicative of a local detection outcome at the first grid-node;
• generate, for the time-step index, first feature opinion information for the area based on sensor information captured by the first robot along the Markovian random walk by normalizing the occupancy vector to obtain a first occupancy distribution over the grid-node identifiers;
• access, via the local inter-robot network interface and when the first robot and a second robot of the plurality of robots are co-located at the first grid-node, second feature opinion information from the second robot, the second feature opinion information comprising a second occupancy distribution over the grid-node identifiers;
• generate a fused occupancy distribution by weighted element-wise combination of the first occupancy distribution and the second occupancy distribution;
• and
• update, at the processor of the first robot, the first feature opinion information of the first robot for a subsequent time-step index based on the fused occupancy distribution.