Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning-Based Safety-Aware Task Scheduling for Efficient Human-Robot Collaboration

M. Faroni, A. Spano, A. M. Zanchettin, P. Rocco

TL;DR

The paper tackles the efficiency-safety trade-off in human-robot collaboration by learning the mapping from system states to safety-induced speed reductions without needing explicit safety logic. It introduces an Exo-MDP formulation and a two-phase framework: a learning pipeline that predicts average slowdown, and an execution pipeline with greedy and Monte Carlo action selection. Key contributions include a neural-network architecture tailored for staircase safety functions, an online planning approach that minimizes slowdown over a horizon, and validation in both simulated pick-and-pack and a real UR5e setup showing reduced cycle times and robust performance. The work enables safer, more efficient HRI by leveraging data-driven predictions of safety effects rather than relying on detailed, hard-to-obtain safety models. Practical impact is demonstrated through substantial improvements in task timing and smoother human-robot collaboration in structured packaging tasks.

Abstract

Ensuring human safety in collaborative robotics can compromise efficiency because traditional safety measures increase robot cycle time when human interaction is frequent. This paper proposes a safety-aware approach to mitigate efficiency losses without assuming prior knowledge of safety logic. Using a deep-learning model, the robot learns the relationship between system state and safety-induced speed reductions based on execution data. Our framework does not explicitly predict human motions but directly models the interaction effects on robot speed, simplifying implementation and enhancing generalizability to different safety logics. At runtime, the learned model optimizes task selection to minimize cycle time while adhering to safety requirements. Experiments on a pick-and-packaging scenario demonstrated significant reductions in cycle times.

Learning-Based Safety-Aware Task Scheduling for Efficient Human-Robot Collaboration

TL;DR

The paper tackles the efficiency-safety trade-off in human-robot collaboration by learning the mapping from system states to safety-induced speed reductions without needing explicit safety logic. It introduces an Exo-MDP formulation and a two-phase framework: a learning pipeline that predicts average slowdown, and an execution pipeline with greedy and Monte Carlo action selection. Key contributions include a neural-network architecture tailored for staircase safety functions, an online planning approach that minimizes slowdown over a horizon, and validation in both simulated pick-and-pack and a real UR5e setup showing reduced cycle times and robust performance. The work enables safer, more efficient HRI by leveraging data-driven predictions of safety effects rather than relying on detailed, hard-to-obtain safety models. Practical impact is demonstrated through substantial improvements in task timing and smoother human-robot collaboration in structured packaging tasks.

Abstract

Ensuring human safety in collaborative robotics can compromise efficiency because traditional safety measures increase robot cycle time when human interaction is frequent. This paper proposes a safety-aware approach to mitigate efficiency losses without assuming prior knowledge of safety logic. Using a deep-learning model, the robot learns the relationship between system state and safety-induced speed reductions based on execution data. Our framework does not explicitly predict human motions but directly models the interaction effects on robot speed, simplifying implementation and enhancing generalizability to different safety logics. At runtime, the learned model optimizes task selection to minimize cycle time while adhering to safety requirements. Experiments on a pick-and-packaging scenario demonstrated significant reductions in cycle times.

Paper Structure

This paper contains 19 sections, 11 equations, 7 figures, 3 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Related works
  3. Problem Statement
  4. Method
  5. Learning pipeline
  6. Execution pipeline
  7. Greedy Action Selection
  8. Monte Carlo Action Selection
  9. Experiments
  10. Simulated pick&packaging scenario
  11. Data Collection and Training
  12. Learning Results
  13. Execution Results
  14. Real scenario
  15. Data Collection and Training
  16. ...and 4 more sections

Figures (7)

  • Figure 1: Working principle of the proposed approach.
  • Figure 2: Neural network architecture.
  • Figure 3: Simulation scenario.
  • Figure 4: Example of safety speed scaling function for $K=5$.
  • Figure 5: Actual v. predicted average scaling values on test datasets. The heatmap represents the density of samples. A high density along the diagonal denotes a high regression accuracy.
  • ...and 2 more figures