Modular Multi-Level Replanning TAMP Framework for Dynamic Environment

TL;DR

The paper tackles the fragility of traditional TAMP in dynamic environments by introducing the Modular Multi-Level Replanning Framework (MMRF), which fuses a probabilistically complete TAMP solver with real-time, multi-level replanning at the logic and motion levels. By generating a nominal plan $P_n$ and online reconstructing an actual plan $P_a$ through a Subtask Scheduler and verifying feasibility with a Subtask Planner, MMRF reduces the need for expensive full replanning and improves responsiveness to interference. Real-world experiments on stack and rearrange tasks with a Franka Panda demonstrate 100% success across interference scenarios and support notable reductions in completion time (average ~13%, up to ~28% under heavy interference) due to online optimization and parallel motion planning. The framework’s modular design enables integration with diverse TAMP and motion planners and points to future enhancements such as visual-language-model-based logic parsing and optimized scheduling for shortest task paths.

Abstract

Task and Motion Planning (TAMP) algorithms can generate plans that combine logic and motion aspects for robots. However, these plans are sensitive to interference and control errors. To make TAMP more applicable in real-world, we propose the modular multi-level replanning TAMP framework(MMRF), blending the probabilistic completeness of sampling-based TAMP algorithm with the robustness of reactive replanning. MMRF generates an nominal plan from the initial state, then dynamically reconstructs this nominal plan in real-time, reorders robot manipulations. Following the logic-level adjustment, GMRF will try to replan a new motion path to ensure the updated plan is feasible at the motion level. Finally, we conducted real-world experiments involving stack and rearrange task domains. The result demonstrate MMRF's ability to swiftly complete tasks in scenarios with varying degrees of interference.

Figures (7)

• Figure 1: Three levels of interference. (a) Slight interference does not affect the execution order of the plan, but requires motion replan for placing. (b) Middle interference disrupts the sequence of plan. Robot needs to re-pick and re-place the green block. (c) Heavy interference make the plan invalid ,and can only be resolved by TAMP replanning
• Figure 2: Structure of the modular multi-Level replanning TAMP framework. (a) TAMP Solver:generates nominal plans. (b) Subtask Scheduler:Reconstructs the nominal plan into the actual plan to ensure the feasibility at the logic level. (c) Subtask Planner: performs motion planning for the subtasks in the actual plan to ensure the feasibility of the plan at the motion level. (d) Robot Controller: generates control sequences based on the first task in the plan and the current state. (e) State Evaluator: Evaluates numerical parameters based on sensor information and applies predicate resolution rules to evaluate logic state.
• Figure 3: The workflow of the Subtask Planner. First generate the target state and then plan the robot motion path and obtain the end state. (a) place subtask. (b) push subtasks
• Figure 4: The execution of the pick subtask failed due to the sudden obstruction of the target block. However, this subtask is still feasible at the motion level. After repeated execution, the yellow block was successfully grabbed.
• Figure 5: The workflow of the State Evaluator. The State Evaluator first extracts the numerical state based on the sensors and then applies predicate parsing rules to parse out the logic state.