GP-guided MPPI for Efficient Navigation in Complex Unknown Cluttered Environments (2307.04019v3)

Abstract: Robotic navigation in unknown, cluttered environments with limited sensing capabilities poses significant challenges in robotics. Local trajectory optimization methods, such as Model Predictive Path Intergal (MPPI), are a promising solution to this challenge. However, global guidance is required to ensure effective navigation, especially when encountering challenging environmental conditions or navigating beyond the planning horizon. This study presents the GP-MPPI, an online learning-based control strategy that integrates MPPI with a local perception model based on Sparse Gaussian Process (SGP). The key idea is to leverage the learning capability of SGP to construct a variance (uncertainty) surface, which enables the robot to learn about the navigable space surrounding it, identify a set of suggested subgoals, and ultimately recommend the optimal subgoal that minimizes a predefined cost function to the local MPPI planner. Afterward, MPPI computes the optimal control sequence that satisfies the robot and collision avoidance constraints. Such an approach eliminates the necessity of a global map of the environment or an offline training process. We validate the efficiency and robustness of our proposed control strategy through both simulated and real-world experiments of 2D autonomous navigation tasks in complex unknown environments, demonstrating its superiority in guiding the robot safely towards its desired goal while avoiding obstacles and escaping entrapment in local minima. The GPU implementation of GP-MPPI, including the supplementary video, is available at https://github.com/IhabMohamed/GP-MPPI.

GP-guided MPPI for Efficient Navigation in Complex Unknown Cluttered Environments

Introduction

This paper presents a novel control strategy for robotic navigation in environments that are both unknown and cluttered. The proposed approach combines Model Predictive Path Integral (MPPI), a local trajectory optimization method, with a Sparse Gaussian Process (SGP) perception model. This paper introduces the GP-MPPI control strategy, which leverages SGPs to construct a variance surface for recommending optimal subgoals without needing a global map or offline training.

Figure 1: Architecture of our proposed GP-MPPI control strategy, which comprises two main components: the GP-subgoal recommender and the local planner, the MPPI.

Methodology

GP-MPPI integrates a GP-based perception model with MPPI to address navigation challenges in environments where sensing is limited. The SGP occupancy model creates a variance surface, providing insights into navigable spaces around the robot. This model informs the GP-subgoal recommender to suggest subgoals that the local MPPI planner uses to guide navigation safely and efficiently.

GP-subgoal Recommender and MPPI

The GP-subgoal recommender evaluates the navigable space using the SGP variance surface, identifying potential subgoals and recommending the optimal one based on a cost function. This recommendation directly informs the MPPI planner, which considers control inputs that minimize predefined cost functions. Through this iterative process, the robot navigates towards its goal while avoiding collisions and escaping local minima.

Figure 2: Illustrative example of the SGP occupancy model, demonstrating the Jackal robot in a cluttered environment and the roles of raw pointcloud data, the SGP occupancy surface, and variance surface.

Simulation Results

Simulated trials in both forest-like and maze-like environments tested the efficacy of GP-MPPI. Results show that GP-MPPI consistently outperformed vanilla MPPI and log-MPPI in navigating complex terrains, maintaining a task completion rate of 100% without getting caught in local minima.

• Forest Scenario: GP-MPPI achieved an average velocity of 1.30 m/s and maintained collision-free navigation, outperforming baseline methods in task completion rates.
• Maze Scenario: Through two mission types, GP-MPPI effectively navigated the environment by leveraging the subgoal recommendations to avoid local minima, with an average distance reduction in recovery mode.

Figure 3: Robot paths in maze-like environments showcasing different strategies of GP-MPPI against local minima issues presented by baseline methods.

Real-World Demonstration

The proposed control strategy was validated in a real-world indoor environment with randomly placed obstacles. The experiments reinforced the simulation findings, demonstrating GP-MPPI's capability