Optimization-Based Trajectory Tracking Approach for Multi-Rotor Aerial Vehicles in Unknown Environments

Geesara Kulathunga, Hany Hamed, Dmitry Devitt, Alexandr Klimchik

TL;DR

This paper introduces an optimization-based trajectory tracking system for multi-rotor aerial vehicles (MAVs) operating in unknown, cluttered environments. It achieves this by employing a two-pronged approach: a global planner that refines the initial trajectory to avoid obstacles and a local planner that generates a safe and dynamically feasible control policy. This method significantly improves success rates in challenging scenarios compared to existing approaches.

What Problem Does This Paper Address?

Safely and efficiently navigating multi-rotor aerial vehicles (MAVs) through unknown, cluttered environments is a significant challenge. Existing trajectory tracking methods often struggle with real-time performance, collision avoidance, and dynamic feasibility, especially when faced with unexpected obstacles. This is particularly true for low-speed maneuvers where precise control is crucial. The paper aims to bridge this gap by developing a robust system that can continuously refine a planned trajectory to steer clear of obstacles while simultaneously generating a safe and executable control command.

Methods

The core of the proposed approach is a parallel architecture combining a global planner and a local planner.

  • Global Planner: This component is responsible for refining the initial reference trajectory. It uses two convex programming techniques:
  • Trajectory Refinement: It identifies segments of the reference trajectory that intersect with known obstacles and "pushes" them towards free space. This involves identifying pushing directions and calculating gradient vectors to adjust control points.
  • Dead Zone Recovery: If the primary refinement fails, a secondary mechanism is employed to push the entire trajectory away from obstacle zones, ensuring a feasible path.
  • Smoothing and Feasibility: The global planner also incorporates objectives to ensure the refined trajectory is dynamically feasible by penalizing excessive velocity and acceleration.
  • Local Planner: This component takes the refined trajectory from the global planner and generates a near-optimal control policy. It uses a constrained nonlinear model predictive control (NMPC) formulation to ensure safety, dynamic feasibility, and accurate tracking of the refined trajectory. The NMPC is designed to handle close-in obstacles and system dynamics within a prediction horizon.

The system utilizes an Octomap for environment mapping and Euclidean Distance Transform Mapping (EDTM) for estimating free space. The global and local planners operate in parallel, with the global planner focusing on real-time performance and obstacle avoidance, and the local planner ensuring safety and dynamic feasibility.

Datasets

The paper does not explicitly mention using pre-existing datasets for training. Instead, it relies on:

  • Simulated Experiments: Performed in a Gazebo environment.
  • Real-World Experiments: Conducted using a DJI M100 MAV equipped with a Velodyne Lite 16 lidar.
  • New Benchmark: The authors introduce a benchmark using 12 randomly generated forest environments to compare their approach against others. These environments are designed to be cluttered and test the system's ability to handle varying densities.

Key Insights and Ideas

  • Parallel Global-Local Planning Synergy: The key insight is the effective synergy between a global planner that proactively avoids obstacles by refining the trajectory and a local planner that ensures safe execution. This parallel operation allows for both long-term planning and immediate reaction.
  • Optimization for Obstacle Avoidance: Using convex programming to "push" trajectories away from obstacles is an efficient way to generate safe paths in unknown environments without resorting to computationally expensive sampling-based methods.
  • Dead Zone Recovery for Robustness: The inclusion of a dead zone recovery mechanism is crucial for handling situations where the primary global planner might fail, particularly in complex or uncertain environments.
  • Real-time Performance with NMPC: Despite the computational demands of NMPC, the authors demonstrate that it can be effectively used for low-speed maneuvers within real-time constraints when combined with an efficient global planner.

Key Results

  • High Success Rate: The proposed approach achieved a success fraction of 0.91 in cluttered environments, significantly outperforming other methods like RRT* (0.41), and existing local/global planners (0.58-0.83).
  • Real-time Performance: The global and local planners achieved mean computation times of approximately 0.06s (15Hz) and 0.05s (20Hz), respectively, on an NVIDIA Jetson Xavier NX computer.
  • Tracking Accuracy: In obstacle-free zones, the tracking accuracy was less than 1m.
  • Outperformance in Cluttered Environments: The approach demonstrated superior performance in cluttered environments compared to sampling-based pathfinding, a local planner, and graph-based pathfinding.

Limitations and Future Work

  • Low-Speed Focus: The current approach is primarily designed for low-speed maneuvers. The authors acknowledge that extending it to high-speed maneuvers is a significant future direction. This would likely involve improvements to the local planner to handle higher velocities and accelerations.
  • Map Construction Precision: The paper notes that map construction can be imprecise due to sensor limitations (e.g., small Field of View of depth sensors) and the computational cost of building detailed maps in cluttered environments. This can impact the global planner's ability to accurately identify obstacle zones.
  • Computational Cost for Complex Scenarios: While real-time performance is achieved, further optimization might be needed for extremely complex and dynamic environments.

The authors explicitly state their intention to extend the work for high-speed maneuvers, focusing on improving the local planner.

Summary updated

Abstract

The goal of this paper is to develop a continuous optimization-based refinement of the reference trajectory to 'push it out' of the obstacle-occupied space in the global phase for Multi-rotor Aerial Vehicles in unknown environments. Our proposed approach comprises two planners: a global planner and a local planner. The global planner refines the initial reference trajectory when the trajectory goes either through an obstacle or near an obstacle and lets the local planner calculate a near-optimal control policy. The global planner comprises two convex programming approaches: the first one helps to refine the reference trajectory, and the second one helps to recover the reference trajectory if the first approach fails to refine. The global planner mainly focuses on real-time performance and obstacles avoidance, whereas the proposed formulation of the constrained nonlinear model predictive control-based local planner ensures safety, dynamic feasibility, and the reference trajectory tracking accuracy for low-speed maneuvers, provided that local and global planners have mean computation times 0.06s (15Hz) and 0.05s (20Hz), respectively, on an NVIDIA Jetson Xavier NX computer. The results of our experiment confirmed that, in cluttered environments, the proposed approach outperformed three other approaches: sampling-based pathfinding followed by trajectory generation, a local planner, and graph-based pathfinding followed by trajectory generation.
BibTeX
@article{kulathunga2022optimization,
  author = {Geesara Kulathunga and Hany Hamed and Dmitry Devitt and Alexandr Klimchik},
  title = {{Optimization-Based Trajectory Tracking Approach for Multi-Rotor Aerial Vehicles in Unknown Environments}},
  journal = {IEEE RA-L},
  year = {2022},
  doi = {10.1109/lra.2022.3151157},
}
Figure