Intent Prediction-Driven Model Predictive Control for UAV Planning and Navigation in Dynamic Environments

TL;DR

The paper addresses safe UAV navigation in dynamic environments by integrating a perception module, an MDP-based intent predictor for dynamic obstacles, and an MPC-based trajectory planner that evaluates multiple intent-driven trajectories. By representing obstacle intents as probability distributions and forecasting trajectories for all intents, the method reduces collision risk even under limited sensing and tracking loss. Results from simulations and physical flights show fewer collisions compared with benchmark planners, with real-time performance and robustness to occlusions and observation gaps. The framework advances autonomous UAV safety for construction-site applications and demonstrates practical viability with open-source tooling and real-time operation.

Abstract

Aerial robots can enhance construction site productivity by autonomously handling inspection and mapping tasks. However, ensuring safe navigation near human workers remains challenging. While navigation in static environments has been well studied, navigating dynamic environments remains open due to challenges in perception and planning. Payload limitations restrict the robots to using cameras with limited fields of view, resulting in unreliable perception and tracking during collision avoidance. Moreover, the rapidly changing conditions of dynamic environments can quickly make the generated optimal trajectory outdated.To address these challenges, this paper presents a comprehensive navigation framework that integrates perception, intent prediction, and planning. Our perception module detects and tracks dynamic obstacles efficiently and handles tracking loss and occlusion during collision avoidance. The proposed intent prediction module employs a Markov Decision Process (MDP) to forecast potential actions of dynamic obstacles with the possible future trajectories. Finally, a novel intent-based planning algorithm, leveraging model predictive control (MPC), is applied to generate navigation trajectories. Simulation and physical experiments demonstrate that our method improves the safety of navigation by achieving the fewest collisions compared to benchmarks.

Figures (7)

• Figure 1: The UAV navigating a dynamic environment using the proposed algorithm. The left figure shows our customized UAV equipped with onboard sensors and the right figure illustrates the UAV avoiding a pedestrian.
• Figure 2: The proposed dynamic environment navigation system framework. Given localization, RGB, and depth image data, the static perception module constructs an occupancy map, while the dynamic perception module detects and tracks dynamic obstacles with out-of-view compensation. With the tracked obstacles' histories, the prediction module anticipates future obstacle trajectories using intent prediction. The intent-based planning algorithm, including trajectory planning and selection, then optimizes and selects the highest-score trajectory. Finally, the selected trajectory is sent to the controller for execution.
• Figure 3: Illustration of unsafe dynamic obstacle avoidance due to tracking loss. (a) The robot initially detects the dynamic obstacle, generating a collision-free trajectory. (b) As the robot maneuvers, the obstacle moves out of the camera view, leading to an unsafe replanned trajectory.
• Figure 4: Illustration of the proposed prediction method. (a) Predicted trajectories (red curves) are generated from possible obstacle intents based on the obstacle's historical trajectory and the environment. (b) The example demonstrates trajectory prediction for the obstacle's right (direction) intent. The final predicted trajectories are represented by the mean of the sampled trajectories, with the trajectory variance indicated by shaded green areas.
• Figure 5: Comparison of generated trajectories from benchmark planners in the same scenario. (a) Experiment scenario in Gazebo. (b) EGO Planner ego_planner failure due to a noisy map. (c) Our planner's trajectory without prediction, resulting in potential collision. (d) Safe trajectory generated by our planner.