Automatic Extension of a Symbolic Mobile Manipulation Skill Set

TL;DR

This work proposes a method that allows an agent to automatically extend its skill set, and thus the abstract description, upon encountering such a situation, and introduces strategies for generalizing from previous experience, completing sequences of key actions and discovering preconditions to ensure the efficiency of the skill sequence exploration scheme.

Abstract

Symbolic planning can provide an intuitive interface for non-expert users to operate autonomous robots by abstracting away much of the low-level programming. However, symbolic planners assume that the initially provided abstract domain and problem descriptions are closed and complete. This means that they are fundamentally unable to adapt to changes in the environment or task that are not captured by the initial description. We propose a method that allows an agent to automatically extend its skill set, and thus the abstract description, upon encountering such a situation. We introduce strategies for generalizing from previous experience, completing sequences of key actions and discovering preconditions to ensure the efficiency of our skill sequence exploration scheme. The resulting system is evaluated in simulation on object rearrangement tasks. Compared to a Monte Carlo Tree Search baseline, our strategies for efficient search have on average a 29% higher success rate at a 68% faster runtime.

Figures (4)

• Figure 1: Simulation environment setup that is used to demonstrate the proposed automatic skill set extension method. The unstructured environment is designed for object rearrangement tasks.
• Figure 2: Overview of the exploration for skill set extension with inputs, proposed algorithm components, a physics simulator and intermediate results.
• Figure 3: Types and entities assigned to them, before and after extending the symbolic description adding new sub-types (bold) for the cube, the cupboard and a newly introduced position sample.
• Figure 4: (left) Runtime comparison of all sequence exploration methods for the scenarios shown in Table \ref{['tab:expscenarios']}. Each method was run 20 times for each scenario, with a time budget of 900 seconds. The numbers at the top of the diagram indicate how many of the 20 trials timed out before finding a feasible solution (lower is better). Only the times taken by successful runs are included in the plot. (right) Average simulation time as a percentage of the total planning time for each method and all experiment scenarios.