CSubBT: A Self-Adjusting Execution Framework for Mobile Manipulation System

TL;DR

This paper tackles the challenge of robust long-horizon mobile manipulation under perceptual disturbances by introducing CSubBT, a self-adjusting execution framework based on Behavior Trees. It formalizes symbolic actions as factored transition systems and uses conditional samplers within a centralized CSNode to explore the constraint space and reparameterize atomic actions on-the-fly, avoiding frequent replanning. The approach decomposes actions into CANodes and manages anomaly handling through a shared status table, enabling reactive, modular execution. Empirical results in simulation and real-world tests show high robustness (e.g., 92.5% success in grasping under bias) and favorable reactivity compared to alternative reactive planners, highlighting practical benefits for autonomous mobile manipulation in unstructured environments.

Abstract

With the advancements in modern intelligent technologies, mobile robots equipped with manipulators are increasingly operating in unstructured environments. These robots can plan sequences of actions for long-horizon tasks based on perceived information. However, in practice, the planned actions often fail due to discrepancies between the perceptual information used for planning and the actual conditions. In this paper, we introduce the {\itshape Conditional Subtree} (CSubBT), a general self-adjusting execution framework for mobile manipulation tasks based on Behavior Trees (BTs). CSubBT decomposes symbolic action into sub-actions and uses BTs to control their execution, addressing any potential anomalies during the process. CSubBT treats common anomalies as constraint non-satisfaction problems and continuously guides the robot in performing tasks by sampling new action parameters in the constraint space when anomalies are detected. We demonstrate the robustness of our framework through extensive manipulation experiments on different platforms, both in simulation and real-world settings.

Figures (6)

• Figure 1: A general structure of CSubBT. The execution framework is based on BTs. CSNode collects and handles error information uniformly, then transmits adjustment parameters and the boolean state to each CANode through the blackboard to address errors. More details are in Section \ref{['section: CS Node and CA Node']}.
• Figure 2: The constraint network for the Move-and-Pick process.
• Figure 3: The examples demonstrate how CSubBT handles two common exceptions. The objective of the mobile manipulator is to collect all the cubes on the desk. Error information and newly adjusted input data are transmitted via the blackboard. In example (a), when the robot fails to find a cube by swinging its arm in Cartesian space, the CSNode updates the status table's value (0100) to identify a new base pose that can address the situation. In example (b), the robot lacks a valid trajectory when performing the Pre-approach action with the top grasp mode. The CSNode generates an alternative grasp mode (e.g., forward grasp) to handle this exception. The dashed box in the figure illustrates the equivalent BT of CSubBT for handling the current exception, designed based on traditional BT design principles (PPA and Backchaining). BTs that are fully equivalent to the CSubBT tend to have a more complex structure.
• Figure 4: Gazebo simulation experiment scenarios using a mobile manipulation platform are presented. (a) shows the initial state of the classic mobile grasping experiment. (b) depicts the three cubes in the task, each with varying degrees of perceptual bias. The small green cube indicates the robot's expected position, while the small blue square represents the actual position. (c) illustrates the initial state of the capture occlusion experiment. (d)-(g) show parts of the grasping process for the cubes, the scripts nodes in the Groot2 monitor window of figure (g) are the initial parameters in the blackboard.
• Figure 5: Reactivity experiments: average runtime of the methods over 10 trials per problem size.