SwarmCCO: Probabilistic Reactive Collision Avoidance for Quadrotor Swarms Under Uncertainty

Arul, Senthil Hariharan; Manocha, Dinesh · 2021 · Crossref

DOI: 10.1109/lra.2021.3061975

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of decentralized collision avoidance for quadrotor swarms operating under uncertain state estimation. While prior methods often rely on deterministic assumptions or conservative bounding volume expansions that fail in noisy, real-world environments, this work introduces SwarmCCO, a probabilistic approach that handles uncertainty directly. The motivation stems from the need for scalable, robust navigation algorithms for UAV swarms in uncontrolled outdoor settings where onboard sensors introduce noise in position and velocity estimates. The proposed method exploits the differential flatness property of quadrotors to linearize their non-linear dynamics via feedforward linearization, enabling efficient computation within a Model Predictive Control (MPC) framework. SwarmCCO formulates reciprocal collision avoidance constraints as chance constraints, ensuring that trajectories avoid collisions with a specified confidence level. The authors present two distinct formulations for solving these constraints: Method I assumes a Gaussian noise distribution, reformulating chance constraints into deterministic second-order cone constraints. Method II extends this to non-Gaussian noise by approximating the uncertainty distribution using a Gaussian Mixture Model (GMM) fitted to state samples, splitting collision constraints into multiple second-order cones corresponding to each Gaussian component. The algorithms were evaluated in simulations involving multiple quadrotor agents, comparing performance against deterministic baselines in terms of path length, time to goal, and collision frequency. Results indicate that both probabilistic methods significantly reduce collisions compared to deterministic approaches. The Gaussian method achieved an average computation time of approximately 5ms per agent, while the non-Gaussian method required approximately 9ms in scenarios with four agents. Although the non-Gaussian method is computationally more expensive, it demonstrated superior performance in constricted regions, yielding shorter path lengths and better satisfaction of collision avoidance constraints by more accurately approximating complex noise distributions where the Gaussian method might yield infeasible solutions. The significance of this work lies in providing a scalable, decentralized solution for multi-agent navigation that explicitly accounts for sensor uncertainty without resorting to overly conservative safety margins. By integrating chance-constrained optimization with differential flatness, SwarmCCO enables reliable operation in dynamic, noisy environments. The study highlights the trade-off between computational cost and performance accuracy, suggesting that non-Gaussian modeling is beneficial for complex scenarios despite higher processing demands. This contributes to the field of robotics by advancing the reliability of autonomous swarm systems in practical, real-world applications where perfect state estimation is unavailable.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• time to collision looming