TL;DR
This paper introduces CoNi-OA, a novel obstacle avoidance algorithm for quadrotors operating in a non-inertial frame relative to a ground robot (UGV). It enables safe, cooperative flight by generating collision-free trajectories using only single-frame LiDAR data, without needing global state estimation or explicit obstacle prediction, making it highly adaptable to dynamic and unknown environments.
What Problem Does This Paper Address?
Cooperative missions involving Unmanned Aerial Vehicles (UAVs) and Unmanned Ground Vehicles (UGVs) offer significant advantages in various applications like surveillance and search-and-rescue. A prior work, CoNi-MPC, established an efficient framework for such cooperation by controlling the UAV in the UGV's non-inertial frame, eliminating the need for complex global state estimation. However, this approach lacked a robust mechanism for obstacle avoidance, severely limiting its practical deployment in real-world scenarios where dynamic obstacles are common. The core challenge is enabling the UAV to safely navigate and perform its tasks while accounting for obstacles that appear "dynamic" and unpredictable from its non-inertial perspective, all while adhering to its own dynamic constraints.
Methods
The proposed solution, Cooperative Non-inertial frame-based Obstacle Avoidance (CoNi-OA), tackles this by leveraging a sample-based modulation technique. Here's a breakdown of the key methods:
- Non-Inertial Frame Control: The system operates entirely within the UGV's non-inertial frame. This simplifies control by removing the need for global positioning systems (GPS) or Simultaneous Localization and Mapping (SLAM), making it suitable for featureless or unknown environments.
- Sampled-Based Modulation Matrix: CoNi-OA uses raw LiDAR data from a single frame to construct a modulation matrix. This matrix directly influences the quadrotor's velocity, guiding it away from obstacles. This approach avoids the need for explicit obstacle modeling or prediction.
- Iterative Trajectory Generation: Instead of planning a global path, CoNi-OA iteratively generates local, collision-free trajectories. This is achieved by modulating an initial velocity (provided by a high-level planner) using the derived modulation matrix and the quadrotor's non-inertial dynamics.
- Quadrotor Dynamics Adherence: The generated trajectories are designed to be dynamically feasible for the quadrotor, considering its rigid-body dynamics in the non-inertial frame.
- CoNi-MPC Integration: The generated collision-free trajectories are then tracked by a CoNi-MPC controller, which handles the actual thrust and angular velocity commands for the quadrotor.
Datasets
The paper does not explicitly mention training datasets. For evaluation, the authors conducted:
- Real-world experiments: Using a quadrotor and UGV platform (shown in Fig. 6), with relative state estimation provided by a motion capture system and LiDAR data from a Livox Mid-360 sensor.
- Simulations: Using ROS and Rviz for comparison against other methods. The simulations involved varying obstacle densities and UGV motion constraints.
Key Insights and Ideas
- "Dynamic" Obstacles in Non-Inertial Frames: A crucial insight is how static obstacles in the world frame become "dynamic" and unpredictable from the UGV's non-inertial frame. CoNi-OA directly addresses this by not relying on prior knowledge of obstacle motion.
- Single-Frame LiDAR for Real-Time Avoidance: The ability to generate safe trajectories from a single LiDAR frame in under 5ms is a significant achievement, enabling real-time responsiveness crucial for dynamic environments.
- Modulation as a General Framework: The paper positions CoNi-OA as a general framework for obstacle avoidance, adaptable to various scenarios beyond just UAV-UGV cooperation, including traditional world-frame navigation with static obstacles.
- Bridging High-Level Planning and Low-Level Control: The modulation approach acts as an effective intermediary, translating high-level desired motions into dynamically feasible, collision-free commands for the quadrotor.
Key Results
- Real-World Demonstrations: CoNi-OA successfully demonstrated stable obstacle avoidance during leader-following, orbit flight, and directional landing tasks in real-world cooperative scenarios.
- Computational Efficiency: The trajectory generation process takes an average of 4.91 ms for a 20-step trajectory, enabling a control loop frequency of over 100 Hz. The controller's average computation time is a mere 0.24 ms.
- Superior Performance in Simulations: In simulations comparing CoNi-OA with Ego-Planner, VO, and AVO, CoNi-OA achieved significantly higher success rates (up to 94% in some scenarios) across various obstacle densities and UGV motion speeds. For instance, in the sparse obstacle, slow UGV movement, leader-follower task, CoNi-OA achieved a 94% success rate, while Ego-Planner achieved 66%, VO 62%, and AVO 61%.
- Robustness to Dynamic Environments: The method proved effective even when the UGV moved unpredictably, creating dynamic obstacles from the UAV's perspective.
Limitations and Future Work
- LiDAR Update Frequency: The authors acknowledge that the relatively low update frequency of the LiDAR sensor can limit the method's ability to generate timely collision-free trajectories, especially during aggressive UGV maneuvers.
- Future Work:
- Integrating higher-frequency LiDAR sensors.
- Extending CoNi-OA to multi-robot cooperative obstacle avoidance.
- Incorporating real-time relative estimation devices to improve accuracy.
Summary updated