PA-MPPI: Perception-Aware Model Predictive Path Integral Control for Quadrotor Navigation in Unknown Environments

TL;DR

PA-MPPI tackles autonomous quadrotor navigation in unknown environments by augmenting Model Predictive Path Integral control with a perception-driven cost that steers trajectory samples toward informative frontiers. The approach tightly integrates a perception and mapping module (ROG-Map) with a ray-tracing frontier mechanism, enabling real-time, reference-free planning at 50 Hz. Hardware and simulated experiments show PA-MPPI performs on par with state-of-the-art planners like SUPER and serves as a robust action policy for navigation foundation models. Overall, the work advances perception-aware optimization for agile, safe navigation in cluttered, partially observed spaces and suggests avenues for longer-horizon planning and integration with learned planning systems.

Abstract

Quadrotor navigation in unknown environments is critical for practical missions such as search-and-rescue. Solving this problem requires addressing three key challenges: path planning in non-convex free space due to obstacles, satisfying quadrotor-specific dynamics and objectives, and exploring unknown regions to expand the map. Recently, the Model Predictive Path Integral (MPPI) method has emerged as a promising solution to the first two challenges. By leveraging sampling-based optimization, it can effectively handle non-convex free space while directly optimizing over the full quadrotor dynamics, enabling the inclusion of quadrotor-specific costs such as energy consumption. However, MPPI has been limited to tracking control that optimizes trajectories only within a small neighborhood around a reference trajectory, as it lacks the ability to explore unknown regions and plan alternative paths when blocked by large obstacles. To address this limitation, we introduce Perception-Aware MPPI (PA-MPPI). In this approach, perception-awareness is characterized by planning and adapting the trajectory online based on perception objectives. Specifically, when the goal is occluded, PA-MPPI incorporates a perception cost that biases trajectories toward those that can observe unknown regions. This expands the mapped traversable space and increases the likelihood of finding alternative paths to the goal. Through hardware experiments, we demonstrate that PA-MPPI, running at 50 Hz, performs on par with the state-of-the-art quadrotor navigation planner for unknown environments in challenging test scenarios. Furthermore, we show that PA-MPPI can serve as a safe and robust action policy for navigation foundation models, which often provide goal poses that are not directly reachable.

Figures (10)

• Figure 1: The PA-MPPI controller navigating to the goal pose while avoiding obstacles in a previously unknown environment. The controller simultaneously controls the quadrotor at 50Hz and optimizes the perception objective based on an online-updated 3D map constructed from onboard observations.
• Figure 2: Illustration of the control stack. The perception and mapping module has three parts: (1) a point cloud is generated from the rendered depth image and transformed to the world frame, (2) the ROG-Map ROG-Map updates the occupancy map with the new point cloud, and (3) the ROG-Map is converted to a JAX array as input to PA-MPPI.
• Figure 3: A top-down visualization of the ray-tracing in perception cost calculation, showing the occupied voxels (red), free voxels (green), and unknown voxels (blue), and sampled trajectories on which ray-tracing is performed.
• Figure 4: Synthetic scenes for navigation experiments. The goal pose for each task is always 3m ahead of the initial pose, with three types of obstacles in between: a C-shaped wall (a), a wall with a hole (b), and four walls (c). The most challenging setting for each scene is illustrated here, with example successful trajectories depicted in blue.
• Figure 5: Although EGO-Planner EGO-Planner is able to optimize the trajectory to navigate past the C-shaped wall in the easy setting (a), it struggles to find a feasible trajectory in the harder settings (b)(c). Yellow/Green: A* obstacle avoidance proposals, Blue: planned trajectories.