TOLD algorithm cuts robot swarm distortion by 62% in real-time

Conventional robot formation control relies on node-level robust controllers that adjust individual robot inputs without modifying the interaction topology between agents. This approach limits adaptability when faced with intermittent disturbances or dynamic environments. A new paper by Saksham Sharma and colleagues tackles this problem with TOLD (Topological Online Learning for Displacement-based formation control), a framework that updates edge-level interaction weights in real time to directly minimize formation distortion. The team proposes two strategies: Online Gradient Flow (OGF), which uses unconstrained weights, and Online Exponential Gradient Flow (OExpGF), which employs non-negative convex weights. Theoretical analysis shows OExpGF guarantees asymptotic consensus for single-integrator agents over directed graphs, while OGF ensures bounded formation distortion.

Simulations with twelve robots under intermittent disturbances demonstrate a median cumulative root mean distortion error reduction ranging from 1.2% to 33.14% when TOLD is combined with existing node-level controllers. Hardware experiments on Crazyflie 2.0 quadrotors deliver even more striking results: OGF reduces median formation distortion by over 62%, and OExpGF by 31.4%, compared to fixed-weight consensus. These results highlight TOLD's potential for robust, real-time formation control in applications like drone swarms for search and rescue, environmental monitoring, and tactical surveillance where maintaining precise relative positioning under disturbances is critical.