DCAD: Decentralized Collision Avoidance With Dynamics Constraints for Agile Quadrotor Swarms

Arul, Senthil Hariharan; Manocha, Dinesh · 2020 · Crossref

DOI: 10.1109/lra.2020.2967281

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 introduces DCAD (Decentralized Collision Avoidance with Dynamics Constraints), a novel algorithm designed to enable safe, high-velocity navigation for large swarms of quadrotors in dense environments containing static and dynamic obstacles. The research addresses the limitations of existing decentralized collision avoidance methods, which often rely on linearizing quadrotor dynamics around a hover point. This approximation fails during aggressive maneuvers requiring large pitch and roll angles, leading to reduced performance or collisions. Additionally, prior methods frequently ignore the destabilizing effect of rotor downwash or model it using simplified axis-aligned shapes that do not account for vehicle orientation. The authors aim to provide a scalable, real-time solution that incorporates full non-linear dynamics and downwash effects while handling sensor uncertainty. The proposed method combines Optimal Reciprocal Collision Avoidance (ORCA) with a flatness-based Model Predictive Control (MPC) framework. Instead of linearizing around equilibrium, the authors use feedforward linearization based on differential flatness to transform the non-linear quadrotor dynamics into a linear model in "flat space" (position, velocity, acceleration, and yaw). This allows the use of efficient linear MPC for trajectory generation while retaining the ability to map control inputs back to the non-linear physical system. To address downwash, the algorithm models neighboring quadrotors as a combination of a sphere and an oriented ellipsoid. Specifically, the quadrotor at a higher altitude is modeled as an oriented ellipsoid rotated according to its attitude, while the lower quadrotor is modeled as a sphere. This approach ensures symmetric velocity obstacle constraints and accurately reflects the instability region caused by downwash. Sensor uncertainty in position and velocity is managed using Kalman filtering. Experimental evaluations were conducted using the PX4 Software-in-the-Loop framework, ROS, and Gazebo simulations on an Intel Xeon processor. The results demonstrate that DCAD computes collision-avoiding control inputs in approximately 5 milliseconds on average for an agent surrounded by eight obstacles. When compared to state-of-the-art decentralized methods such as ORCA, AVO, LQR-obstacles, and LSwarm, DCAD produced smoother trajectories and exhibited a lower probability of collision during high-velocity maneuvers. The algorithm successfully avoided agents entering each other’s downwash regions during close-proximity operations. The significance of this work lies in its ability to bridge the gap between computational efficiency and dynamic fidelity in swarm robotics. By utilizing flatness-based linearization, DCAD avoids the computational expense and local minima issues associated with Non-linear MPC, while overcoming the performance degradation of hover-point linearization during aggressive flight. The incorporation of orientation-aware downwash modeling enhances safety in dense swarm scenarios. These contributions make DCAD a robust solution for applications requiring rapid, coordinated navigation, such as disaster response and search-and-rescue operations, where scalability and real-time adaptability are critical.

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