LF-3PM: a LiDAR-based Framework for Perception-aware Planning with Perturbation-induced Metric

Kaixin Chai, Long Xu, Qianhao Wang, Chao Xu, Peng Yin, Fei Gao

Abstract: Just as humans can become disoriented in featureless deserts or thick fogs, not all environments are conducive to the Localization Accuracy and Stability (LAS) of autonomous robots. This paper introduces an efficient framework designed to enhance LiDAR-based LAS through strategic trajectory generation, known as Perception-aware Planning. Unlike vision-based frameworks, the LiDAR-based requires different considerations due to unique sensor attributes. Our approach focuses on two main aspects: firstly, assessing the impact of LiDAR observations on LAS. We introduce a perturbation-induced metric to provide a comprehensive and reliable evaluation of LiDAR observations. Secondly, we aim to improve motion planning efficiency. By creating a Static Observation Loss Map (SOLM) as an intermediary, we logically separate the time-intensive evaluation and motion planning phases, significantly boosting the planning process. In the experimental section, we demonstrate the effectiveness of the proposed metrics across various scenes and the feature of trajectories guided by different metrics. Ultimately, our framework is tested in a real-world scenario, enabling the robot to actively choose topologies and orientations preferable for localization.

Introduction

Autonomous robots rely heavily on sensors like LiDAR to determine their pose within an environment. However, certain geometric configurations—such as perfectly smooth corridors or wide-open, featureless fields—can lead to "localization degeneracy," where sensor data remains virtually unchanged despite robot movement. Perception-aware Planning (PAP) seeks to mitigate this by guiding robots toward areas that maximize localization accuracy and stability (LAS).

While PAP has been explored for vision-based systems, LiDAR-based systems present unique challenges. Simulating dense LiDAR scans for real-time trajectory optimization is computationally prohibitive, and existing metrics often fail to provide a theoretically grounded evaluation of a location's geometric "localizability."

The LF-3PM Framework

LF-3PM (LiDAR-based Framework for Perception-aware Planning with Perturbation-induced Metric) addresses these hurdles by decoupling the perception evaluation from the motion planning phase. The framework identifies areas of high observation loss, such as the "degenerate" and "unstructured" scenarios shown in Figure 1, and generates a map to guide the robot toward more reliable regions.

Figure 1: Illustration of degenerate (a) and unstructured (b) scenarios, their spatial context (c), and the resulting Static Observation Loss Map (d) where green indicates high localizability and red indicates high observation loss.

Perturbation-induced Metric

The authors treat localization as a Least Squares problem and derive a metric based on sensitivity analysis. This "perturbation-induced metric" quantifies how much a tiny amount of noise in LiDAR data affects the estimated pose. If a small perturbation causes a large jump in the estimated position, that location is deemed to have high observation loss.

Figure 6: Visualization of the perturbation-induced metric, showing the sensitivity of the pose estimation to small translation perturbations.

The framework offers three variations of this metric ($q^{(min)}$, $q^{(n)}$, and $q^{(max)}$). Experimental results suggest that $q^{(n)}$ serves as a "Goldilocks" metric, providing a balanced assessment that outperforms standard degeneracy factors in both structured and unstructured environments.

Static Observation Loss Map (SOLM)

To overcome the computational cost of real-time LiDAR simulation, LF-3PM introduces the Static Observation Loss Map (SOLM). Before the robot begins its mission, the environment is discretized into a grid. At each grid center, a LiDAR scan is simulated once, and the observation loss is calculated and stored. During the planning phase, the robot simply looks up these pre-computed values, turning a heavy simulation task into a light map-lookup problem.

The Planning Pipeline

The planning process is divided into two stages:

  1. Front-end: A Breadth-First Search (BFS) is performed on the SOLM to identify a path with the lowest cumulative observation loss.
  2. Back-end: The trajectory is refined using the MINCO (Minimum Control effort) framework. This optimization balances localization quality with obstacle avoidance and robot dynamics to ensure smooth and feasible motion.

Experimental Validation

The framework was rigorously tested using Unreal Engine 5 simulations and real-world experiments with a RoboMaster AI-2020 omnidirectional robot. The results demonstrated that the $q^{(n)}$ metric has a high correlation with the Mean Disturbance-induced Error (MDE), the ground truth for localization difficulty.

In real-world tests, trajectories generated by LF-3PM resulted in significantly lower localization errors compared to standard paths. The robot demonstrated "active" behavior, such as choosing to move closer to clusters of objects or orienting its sensor toward stable geometric features to "anchor" its position when navigating empty spaces.

Conclusion

LF-3PM provides a robust and efficient solution for LiDAR-based perception-aware planning. By decoupling evaluation from planning via the SOLM and utilizing a sensitivity-based metric, it enables robots to navigate complex environments while maintaining high localization stability. Future work will focus on adapting the SOLM for dynamic environments where obstacles may move or the map may change over time.

Summary updated

Abstract

Just as humans can become disoriented in featureless deserts or thick fogs, not all environments are conducive to the Localization Accuracy and Stability (LAS) of autonomous robots. This paper introduces an efficient framework designed to enhance LiDAR-based LAS through strategic trajectory generation, known as Perception-aware Planning. Unlike vision-based frameworks, the LiDAR-based requires different considerations due to unique sensor attributes. Our approach focuses on two main aspects: firstly, assessing the impact of LiDAR observations on LAS. We introduce a perturbation-induced metric to provide a comprehensive and reliable evaluation of LiDAR observations. Secondly, we aim to improve motion planning efficiency. By creating a Static Observation Loss Map (SOLM) as an intermediary, we logically separate the time-intensive evaluation and motion planning phases, significantly boosting the planning process. In the experimental section, we demonstrate the effectiveness of the proposed metrics across various scenes and the feature of trajectories guided by different metrics. Ultimately, our framework is tested in a real-world scenario, enabling the robot to actively choose topologies and orientations preferable for localization. The source code is accessible at https://github.com/ZJU-FAST-Lab/LF-3PM.
BibTeX
@inproceedings{chai2024lf,
  author = {Kaixin Chai and Long Xu and Qianhao Wang and Chao Xu and Peng Yin and Fei Gao},
  title = {{LF-3PM: a LiDAR-based Framework for Perception-aware Planning with Perturbation-induced Metric}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2024},
  url = {https://openreview.net/forum?id=USz2WepeiP},
}
Figure