DECAF: a Discrete-Event based Collaborative Human-Robot Framework for Furniture Assembly

TL;DR

The paper addresses long-horizon human-robot collaboration in furniture assembly where the human is treated as an uncontrollable agent. It introduces DECAF, a framework that models the task as a Discrete-Event Markov Decision Process ($DE-MDP$) augmented by a Hierarchical Task Model ($HTM$) and Bayesian human-intent inference. DECAF offers two solution paths—a deterministic decision-graph and a reinforcement-learning policy (PPO)—to handle asynchronous, variable-duration actions and unpredictable events. Validation through simulations (including an IKEA chair) and a real user study with 10 participants demonstrates reduced completion times and improved collaboration experience, while highlighting areas for future work in safety and ergonomics.

Abstract

This paper proposes a task planning framework for collaborative Human-Robot scenarios, specifically focused on assembling complex systems such as furniture. The human is characterized as an uncontrollable agent, implying for example that the agent is not bound by a pre-established sequence of actions and instead acts according to its own preferences. Meanwhile, the task planner computes reactively the optimal actions for the collaborative robot to efficiently complete the entire assembly task in the least time possible. We formalize the problem as a Discrete Event Markov Decision Problem (DE-MDP), a comprehensive framework that incorporates a variety of asynchronous behaviors, human change of mind and failure recovery as stochastic events. Although the problem could theoretically be addressed by constructing a graph of all possible actions, such an approach would be constrained by computational limitations. The proposed formulation offers an alternative solution utilizing Reinforcement Learning to derive an optimal policy for the robot. Experiments where conducted both in simulation and on a real system with human subjects assembling a chair in collaboration with a 7-DoF manipulator.

Figures (6)

• Figure 1: Schematic of a complete flow to achieve collaborative furniture assembly. This work focuses on the components highligted in red.
• Figure 2: Experimental setup for the Ivar chair assembly task.
• Figure 3: The HTM of the Ivar chair assembly task.
• Figure 4: Completion time distribution obtained with the real user experiments described in section \ref{['sec:real_experiment']}.
• Figure 5: Nasa TLX questionnaire. Q1: How mentally demanding was the task? Q2: How physically demanding was the task? Q3: How hurried or rush was the pace of the task? Q4: How successfully were you in accomplishing the task? Q5: How hard did you have to work to accomplish the task? Q6: How insecure, discouraged, irritated, stressed, or annoyed were you?