Many robotic applications involve scenarios where multiple robots interact and the decisions of one robot influence the behavior of others. This paper presents a quadratic programming (QP)-based feedback controller for multi-robot conflict resolution with input constraints. Safety is enforced using coupled Control Barrier Function (CBF), while a yielding matrix encodes how robots influence each other. Unlike existing methods with fixed or hard priorities, the proposed framework enables yielding preferences while guaranteeing deadlock-free operation by design. Sufficient conditions are established for the controller to be locally Lipschitz. The optimal CBF sign- indefinite slack is shown to admit a finite uniform bound, which is compensated through an enlarged CBF safety margin to obtain a forward invariant set contained within the desired safe set. The analysis also identifies an irreducible slack component due to finite maximum deceleration that cannot be eliminated by increasing the slack penalty. Finally, sufficient conditions on the yielding matrix and nominal controller are derived to exclude deadlock states. Simulations demonstrate adaptive coordination across yielding preferences and robustness to robot failures.
Figures & tables
Fig. 1 : Two mobile robots, modeled as disks of radius d , navigate narrow corridors, through a conflict zone (green shaded hexagon), where ψm denotes the minimum pairwise intersection angle among the corridors. The unsafe region is a disk with a dynamically changing radius min{R1,R2} , where each Ri equals the robot’s stopping distance based on its velocity vi , plus a desired safety margin. No two robots may occupy an unsafe region simultaneously.
Fig. 2 : (a)-(b): ν=0.9 , robots 1 and 2 yield preferences: q12=2 , q21=1 . Robot distances from the conflict-zone center are shown. The red arrow shows robot 1 exiting a circle of radius d0 centered at the origin of the conflict zone, after which the conflict with robot 2 is resolved. (c)-(d): ν=0.9 , with swapped yield preferences q12=1 , q21=2 , showing robots’ speed trajectories for vr=3[m/s] .
Fig. 3 : (a)-(b): Robots 1 and 2 yield preferences: q12=0.1 , q21=0.05 . (a) For ν=0.1 , the exit time of yielding robot 2 increases compared to Fig. 2 , as yielding reduces its acceleration and, thus speed. (b) ν=0.1 provides greater weight on asymmetric coupling cost, producing a large initial acceleration difference.
Fig. 4 : ν=0.9 , with robots 1 and 2 yield preferences: q12=2 , q21=1 , inducing robot 2 to yield. When robot 1 fails at 10[m] from the origin, it initiates emergency braking um=−3.4[m/s2] .