Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CAT-RRT: Motion Planning that Admits Contact One Link at a Time

Nataliya Nechyporenko, Caleb Escobedo, Shreyas Kadekodi, Alessandro Roncone

TL;DR

The paper tackles the limitation of binary collision checks in motion planning by enabling controlled contact through a per-link cost framework. It introduces CAT-RRT, an optimization-based planner that assigns individual temperatures to each arm link and uses a per-link cost heuristic to admit contact one link at a time, guided by a transition test. Through simulation and real-world demonstrations, CAT-RRT is shown to reach high-cost regions more reliably and to produce shorter trajectories with balanced contact depth compared to baselines like T-RRT, VF-RRT, RRT*, and BIT*. The approach promises practical impact for manipulation in cluttered and unstructured environments by enabling more flexible interaction with the environment while controlling contact costs.

Abstract

Current motion planning approaches rely on binary collision checking to evaluate the validity of a state and thereby dictate where the robot is allowed to move. This approach leaves little room for robots to engage in contact with an object, as is often necessary when operating in densely cluttered spaces. In this work, we propose an alternative method that considers contact states as high-cost states that the robot should avoid but can traverse if necessary to complete a task. More specifically, we introduce Contact Admissible Transition-based Rapidly exploring Random Trees (CAT-RRT), a planner that uses a novel per-link cost heuristic to find a path by traversing high-cost obstacle regions. Through extensive testing, we find that state-of-the-art optimization planners tend to over-explore low-cost states, which leads to slow and inefficient convergence to contact regions. Conversely, CAT-RRT searches both low and high-cost regions simultaneously with an adaptive thresholding mechanism carried out at each robot link. This leads to paths with a balance between efficiency, path length, and contact cost.

CAT-RRT: Motion Planning that Admits Contact One Link at a Time

TL;DR

The paper tackles the limitation of binary collision checks in motion planning by enabling controlled contact through a per-link cost framework. It introduces CAT-RRT, an optimization-based planner that assigns individual temperatures to each arm link and uses a per-link cost heuristic to admit contact one link at a time, guided by a transition test. Through simulation and real-world demonstrations, CAT-RRT is shown to reach high-cost regions more reliably and to produce shorter trajectories with balanced contact depth compared to baselines like T-RRT, VF-RRT, RRT*, and BIT*. The approach promises practical impact for manipulation in cluttered and unstructured environments by enabling more flexible interaction with the environment while controlling contact costs.

Abstract

Current motion planning approaches rely on binary collision checking to evaluate the validity of a state and thereby dictate where the robot is allowed to move. This approach leaves little room for robots to engage in contact with an object, as is often necessary when operating in densely cluttered spaces. In this work, we propose an alternative method that considers contact states as high-cost states that the robot should avoid but can traverse if necessary to complete a task. More specifically, we introduce Contact Admissible Transition-based Rapidly exploring Random Trees (CAT-RRT), a planner that uses a novel per-link cost heuristic to find a path by traversing high-cost obstacle regions. Through extensive testing, we find that state-of-the-art optimization planners tend to over-explore low-cost states, which leads to slow and inefficient convergence to contact regions. Conversely, CAT-RRT searches both low and high-cost regions simultaneously with an adaptive thresholding mechanism carried out at each robot link. This leads to paths with a balance between efficiency, path length, and contact cost.
Paper Structure (22 sections, 7 equations, 8 figures, 2 tables, 1 algorithm)

This paper contains 22 sections, 7 equations, 8 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Problem description
  5. Motion planning with CAT-RRT
  6. Defining the cost heuristics
  7. Controlling cost magnitude
  8. Per-link cost heuristic used with CAT-RRT
  9. Comparison with state of the art
  10. Obstacle overlap heuristic used with T-RRT, RRT*, and BIT*
  11. Upstream Criterion used with VF-RRT
  12. Experiments
  13. Specifications of the experimental testbed
  14. Simulated experimental scenarios
  15. Metrics for evaluation
  16. ...and 7 more sections

Figures (8)

  • Figure 1: CAT-RRT is an optimization planner which uses a per-link cost heuristic to generate a path in clutter by allowing contact to occur if it is necessary to succeed at the task. Rather than invalidating contact states or restricting motion for the entire arm (left), we propose a method that generates a path by prioritizing the least impacted links (right).
  • Figure 2: Example scenario for per-link (top row) and whole-arm cost (bottom row) with a common start configuration (leftmost vignette). Object--robot contact is shown in orange and sampled states are grouped in gray scale depending on when the state was sampled, darker states are sampled later in time. The rightmost column depicts the planning space of both cost heuristics with green being a low cost area and orange being high cost. The per-link cost planner is able to find the goal location due to a reduced high cost area surrounding the goal location even though some links are in contact with the object.
  • Figure 3: The image on the left shows the robot in front of a point cloud of an object sitting on top of a table. The image on the right shows the same scene with an overlay of point obstacles in red and robot control points in green.
  • Figure 4: Robot's initial configuration ($\mathbf{q}_{near}$) in white, goal configuration ($\mathbf{q}_{goal}$) in green, random sampled state ($\mathbf{q}_{rand}$) in purple, and a set of point obstacles ($\Lambda$) in red. The directional vector for link number six, $\vec{\mathbf{d}}_{l_6}$, points from $\mathbf{q}_{near}$ to $\mathbf{q}_{rand}$. The desired directional vector for the link, $\vec{\mathbf{v}}_{l_6}$, is a weighted sum between the vector from $\Lambda$ to $\mathbf{q}_{near}$ and the vector from $\mathbf{q}_{near}$ to $\mathbf{q}_{goal}$.
  • Figure 5: In our evaluation, the robot is tasked with finding a path from the start state (white) to the goal state (green) while moving through obstacle regions (red) in four experimental scenarios of increasing complexity. The scenarios from left to right are increasingly more complex with obstacles overlapping with the start and goal states.
  • ...and 3 more figures