Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Inductive Learning of Robot Task Knowledge from Raw Data and Online Expert Feedback

Daniele Meli, Paolo Fiorini

TL;DR

This work tackles the challenge of interpretable, trustworthy robotic cognition by learning task knowledge from raw video-kinematic data using inductive logic programming under Answer Set Programming, with temporal reasoning via the event calculus. It introduces a data-efficient offline pipeline to acquire action preconditions, constraints, and effects, and addresses noise through context-dependent partial interpretations learned with ILASP. An online refinement framework with human-in-the-loop ensures safe execution and continuous improvement of the learned knowledge. Evaluations on peg transfer and robotic surgery benchmarks demonstrate robustness and data- and time-efficiency, indicating potential scalability to more complex domains and real-world safety guarantees through interpretable planning.

Abstract

The increasing level of autonomy of robots poses challenges of trust and social acceptance, especially in human-robot interaction scenarios. This requires an interpretable implementation of robotic cognitive capabilities, possibly based on formal methods as logics for the definition of task specifications. However, prior knowledge is often unavailable in complex realistic scenarios. In this paper, we propose an offline algorithm based on inductive logic programming from noisy examples to extract task specifications (i.e., action preconditions, constraints and effects) directly from raw data of few heterogeneous (i.e., not repetitive) robotic executions. Our algorithm leverages on the output of any unsupervised action identification algorithm from video-kinematic recordings. Combining it with the definition of very basic, almost task-agnostic, commonsense concepts about the environment, which contribute to the interpretability of our methodology, we are able to learn logical axioms encoding preconditions of actions, as well as their effects in the event calculus paradigm. Since the quality of learned specifications depends mainly on the accuracy of the action identification algorithm, we also propose an online framework for incremental refinement of task knowledge from user feedback, guaranteeing safe execution. Results in a standard manipulation task and benchmark for user training in the safety-critical surgical robotic scenario, show the robustness, data- and time-efficiency of our methodology, with promising results towards the scalability in more complex domains.

Inductive Learning of Robot Task Knowledge from Raw Data and Online Expert Feedback

TL;DR

This work tackles the challenge of interpretable, trustworthy robotic cognition by learning task knowledge from raw video-kinematic data using inductive logic programming under Answer Set Programming, with temporal reasoning via the event calculus. It introduces a data-efficient offline pipeline to acquire action preconditions, constraints, and effects, and addresses noise through context-dependent partial interpretations learned with ILASP. An online refinement framework with human-in-the-loop ensures safe execution and continuous improvement of the learned knowledge. Evaluations on peg transfer and robotic surgery benchmarks demonstrate robustness and data- and time-efficiency, indicating potential scalability to more complex domains and real-world safety guarantees through interpretable planning.

Abstract

The increasing level of autonomy of robots poses challenges of trust and social acceptance, especially in human-robot interaction scenarios. This requires an interpretable implementation of robotic cognitive capabilities, possibly based on formal methods as logics for the definition of task specifications. However, prior knowledge is often unavailable in complex realistic scenarios. In this paper, we propose an offline algorithm based on inductive logic programming from noisy examples to extract task specifications (i.e., action preconditions, constraints and effects) directly from raw data of few heterogeneous (i.e., not repetitive) robotic executions. Our algorithm leverages on the output of any unsupervised action identification algorithm from video-kinematic recordings. Combining it with the definition of very basic, almost task-agnostic, commonsense concepts about the environment, which contribute to the interpretability of our methodology, we are able to learn logical axioms encoding preconditions of actions, as well as their effects in the event calculus paradigm. Since the quality of learned specifications depends mainly on the accuracy of the action identification algorithm, we also propose an online framework for incremental refinement of task knowledge from user feedback, guaranteeing safe execution. Results in a standard manipulation task and benchmark for user training in the safety-critical surgical robotic scenario, show the robustness, data- and time-efficiency of our methodology, with promising results towards the scalability in more complex domains.
Paper Structure (31 sections, 32 equations, 3 figures, 3 tables, 1 algorithm)

This paper contains 31 sections, 32 equations, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Background and notation
  4. Notation
  5. The peg transfer task
  6. ASP planning task description
  7. Inductive learning of ASP theories
  8. Methodology
  9. Unsupervised learning of ASP task knowledge
  10. Action-context tuples generation
  11. Constraints as extended action preconditions
  12. Learning extended action preconditions
  13. Generating effects of actions
  14. Online supervised refinement of ASP task knowledge
  15. Results
  16. ...and 16 more sections

Figures (3)

  • Figure 1: The setup for the peg transfer task with dVRK.
  • Figure 2: Initial environmental conditions for the task executions used for unsupervised learning.
  • Figure 3: Initial environmental conditions for the task simulations used for task knowledge refinement.

Theorems & Definitions (5)

  • Definition 1: Partial interpretation
  • Definition 2: Context-dependent partial interpretation (CDPI)
  • Definition 3: Learning ASP theories from CDPIs
  • Definition 4: Learning ASP theories from noisy CDPIs
  • Definition 5: Learning ASP theories from positive and negative CDPIs