Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Accurate Pose Prediction on Signed Distance Fields for Mobile Ground Robots in Rough Terrain

Martin Oehler, Oskar von Stryk

TL;DR

The paper addresses accurate prediction of robot-terrain interaction for mobile ground robots with active joints in unstructured environments by introducing an ESDF-based iterative geometric method that estimates the SE(3) pose from an SE(2) pose and joint configuration. The approach uses a Falling Stage to establish contact and a Rotation Stage to refine the pose based on stability margins derived from contact points, leveraging voxel-based surface distances ${\Phi}(\mathbf{p})$ for sub-voxel accuracy. Evaluations in simulation and on a real robot show that the ESDF-based method achieves higher pose-prediction accuracy than heightmaps, especially on rough terrains, with online-planning runtimes on consumer CPUs. The method exhibits robust generalization across two different tracked platforms and RoboCup arenas, and the authors provide open-source ROS implementations and datasets to facilitate reproduction.

Abstract

Autonomous locomotion for mobile ground robots in unstructured environments such as waypoint navigation or flipper control requires a sufficiently accurate prediction of the robot-terrain interaction. Heuristics like occupancy grids or traversability maps are widely used but limit actions available to robots with active flippers as joint positions are not taken into account. We present a novel iterative geometric method to predict the 3D pose of mobile ground robots with active flippers on uneven ground with high accuracy and online planning capabilities. This is achieved by utilizing the ability of signed distance fields to represent surfaces with sub-voxel accuracy. The effectiveness of the presented approach is demonstrated on two different tracked robots in simulation and on a real platform. Compared to a tracking system as ground truth, our method predicts the robot position and orientation with an average accuracy of 3.11 cm and 3.91°, outperforming a recent heightmap-based approach. The implementation is made available as an open-source ROS package.

Accurate Pose Prediction on Signed Distance Fields for Mobile Ground Robots in Rough Terrain

TL;DR

The paper addresses accurate prediction of robot-terrain interaction for mobile ground robots with active joints in unstructured environments by introducing an ESDF-based iterative geometric method that estimates the SE(3) pose from an SE(2) pose and joint configuration. The approach uses a Falling Stage to establish contact and a Rotation Stage to refine the pose based on stability margins derived from contact points, leveraging voxel-based surface distances for sub-voxel accuracy. Evaluations in simulation and on a real robot show that the ESDF-based method achieves higher pose-prediction accuracy than heightmaps, especially on rough terrains, with online-planning runtimes on consumer CPUs. The method exhibits robust generalization across two different tracked platforms and RoboCup arenas, and the authors provide open-source ROS implementations and datasets to facilitate reproduction.

Abstract

Autonomous locomotion for mobile ground robots in unstructured environments such as waypoint navigation or flipper control requires a sufficiently accurate prediction of the robot-terrain interaction. Heuristics like occupancy grids or traversability maps are widely used but limit actions available to robots with active flippers as joint positions are not taken into account. We present a novel iterative geometric method to predict the 3D pose of mobile ground robots with active flippers on uneven ground with high accuracy and online planning capabilities. This is achieved by utilizing the ability of signed distance fields to represent surfaces with sub-voxel accuracy. The effectiveness of the presented approach is demonstrated on two different tracked robots in simulation and on a real platform. Compared to a tracking system as ground truth, our method predicts the robot position and orientation with an average accuracy of 3.11 cm and 3.91°, outperforming a recent heightmap-based approach. The implementation is made available as an open-source ROS package.
Paper Structure (11 sections, 11 equations, 5 figures, 2 tables)

This paper contains 11 sections, 11 equations, 5 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. METHOD
  4. Falling Stage
  5. Rotation Stage
  6. EVALUATION
  7. Simulation
  8. Motion Capture System
  9. Performance
  10. RoboCup rrl 2023 Bordeaux
  11. CONCLUSIONS

Figures (5)

  • Figure 1: Based on an SE(2) pose, the joint configuration and an esdf of the environment (top-left), the 3D pose and terrain interaction are predicted (bottom-left). The photo on the right shows the robot Asterix on the same terrain for comparison.
  • Figure 2: Overview of the pose prediction algorithm. The input pose is used by the Falling Stage to find the first contact with the ground. The Rotation Stage repeatedly rotates the robot about the least stable axis until a stable state is found. Contact candidates are visualized as orange points and the support polygon as green points.
  • Figure 3: Side-view of the Rotation Stage. The rotation axis $\mathbf{r}$, the contact candidate $\mathbf{p}_i$ and the predicted contact point $\mathbf{\hat{c}}_i$ form a triangle. $d_i$ is the distance to the closest surface, retrieved from the esdf. Rotating the robot by $\alpha_i$ would result in a contact.
  • Figure 4: Tracked robot platforms used for evaluation.
  • Figure 5: Overview of the evaluation scenarios in the Gazebo simulator and the DRZ Living Lab. All arenas are part of the RoboCup rrl competition.