[paper-review] Reactive Base Control for On-The-Move Mobile Manipulation in Dynamic Environments

RA-L. [Paper] [Project Page]

Ben Burgess-Limerick^{1, 2}, Jesse Haviland^{1, 2}, Chris Lehnert¹, Peter Corke^{1 1}Queensland University of Technology Centre for Robotics (QCR), Brisbane, Australia ² CSIRO Data61, Brisbane, Australia

Mar. 03

Fig. 1: Overview of Reactive Base Control.

Summary

• This paper presents a reactive-based control method for mobile manipulation in dynamic environments, which significantly reduces the task time and improves the performance of the robot while avoiding static and dynamic obstacles.
• This approach contrasts with traditional methods where the base and manipulator are controlled separately, often resulting in increased task time due to the sequential completion of navigation and manipulation tasks​.

Fig. 2: Inclusion of orientation cost in STAA*.

Fig. 3: Bezier path evaluation with PathRTR.

Contribution

• The paper introduces a reactive base control system for mobile manipulation in dynamic environments, enabling tasks to be performed on-the-move and significantly reducing total task time. This work marks the first real-world demonstration of reactive manipulation with both static and dynamic obstacles, showcasing the method’s practical applicability and robustness. ​
  • 1) They used STAA* algorithm that plangs a collision-free global path by searching through a visibility graph, and then compuites the intersection of the global path and the borader of a local planning region to develop an intermediate goal. The most important addtion to STAA* is the inclusion of an orientation to the goal state. They consider orientation which enables poses to be achieved that smoothly connects the immediate target with the next goal.
  • 2) In A* graph search module, they employed PathRTR heuristic. And they combines it with an additional heuristic based on a Bezier curve. It encourages the exploration of states that can be connected to the goal through smooth curves. They said that a value of 25% results in curves that work well in experimental setup.
  • 3) Base Placement was optimized by simple cost functions by the weighted sum of two components:
    • \[C_i = C_{i,C} + 1.05 \cdot C_{i,N}\]
      • \(C_{i,C}\) is the estimated cost from robot to candidate.
      • \(C_{i,N}\) is the estimated cost from the candidate to the next target.
    • The path cost for each candidate is evaluated using the PathRTR metric.
  • 4) Arm Obstacle Avoidance: They leveraged Quadratic Program to calculate joint veloicited for a given desired end-effector and base velocity. The controller allows for slack in the achieved end-effector velocity.

Thought

• I was already aware of a previous paper by this author, Manipulation On-The-Move, and it seems to be a follow-up to both this paper and an algorithm called STAA*.
• I think the contribution of this paper is to add a couple of heuristics to the existing algorithm, and I would have to start with the previous paper; the baseline controller implemented in this paper is all based on the previous paper as a reference.
  • STAA* / MotM / PathRTR
• After reading the paper and watching the demo video, I was quite impressed with the holistic approach to motion planning for mobile-manipulator systems; so accurate and high performance.