TurboADMM: A Structure-Exploiting Parallel Solver for Multi-Agent Trajectory Optimization
New specialized QP solver achieves near-linear complexity with agent count, solving large coupled optimization problems at control rates.
Researcher Yucheng Chen developed TurboADMM, a specialized QP solver for multi-agent trajectory optimization. It combines three techniques: ADMM decomposition for parallel per-agent solving, Riccati warmstart for temporal structure exploitation, and parametric QP hotstart for reusing KKT factorizations. This enables near-linear scaling with agent count, allowing robots and autonomous vehicles to solve complex, densely-coupled trajectory problems much faster than general-purpose solvers like OSQP or MOSEK.
Why It Matters
Enables real-time coordination for fleets of robots, drones, and autonomous vehicles at previously impossible scales.