Intelligent collision detection and avoidance techniques for autonomous agents
MetadataShow full metadata
Collision is one of the main problems in distributed task cooperation involving multiple moving agents or robots. The collision avoidance problem arises when the environment is dynamic and to reach their destination agents need to use paths that conflict with other agents' paths on specific moves. Decentralized collision avoidance in these situations is more challenging than centralized collision avoidance since autonomous agents must manage their moves independently and may have only a limited capability (local view) to detect the potential risk of collision. Moreover, for true autonomy, there must be no communication between agents or with a central coordinator. This thesis describes novel extensions to current approaches for dealing with collision avoidance and proposes a new dynamic rectangular roundabout (‘rectabout’) collision avoidance method based on human behaviour. The method uses Minimum Enclosing Rectangles (MERs) as potential roundabout carriers to form virtual rectabouts that allow each agent to re-plan its path autonomously and with no communication. This maneuver is calculated independently by each agent involved in a possible collision. The virtual rectabout lies in the intersecting and conflicting position of two agent routes. The approach does not depend on priority schemes and instead involves only local views. MER in turn consists of two components: Minimal Predicted Distance (MPD) detection and MER rectabout collision avoidance algorithm. The MPD is a metric inspired by real human pedestrian collision avoidance behaviour. We use MPD to detect the possible collisions along agent paths and trajectories. The agents involved in conflict will compute a rectabout and re-plan a new velocity when MPD is below the threshold. Experimental simulations involving multi-agent systems indicate that the proposed approach ensures that all agents remain free of collision while attempting to follow their goal direction. The decentralized collision avoidance approach is also applied for WowWee Rovio mobile robots and provides both analytic and empirical evidence to show that the approach generates collision-free motions.