Corrigible Assistance in One Round: Pragmatic-Pedagogic Best Response

2026-07-29

AI Summary

This work addresses the challenge of inefficient robotic assistance in human-robot collaboration caused by asymmetric information about human goals. To overcome the limitations of traditional inverse optimal control in goal disambiguation, the authors propose a pragmatic pedagogical reasoning–based assistive game framework. By incorporating a pedagogical best-response mechanism within a single interaction round, the approach enables rapid and accurate inference of human intent. The method leverages tractable game-theoretic modeling and an efficient solution algorithm to transform the otherwise intractable partially observable Markov decision process (POMDP) into an exactly solvable form. Experimental results in a collaborative block-stacking task demonstrate that the proposed framework accurately and efficiently aligns with human intentions, significantly outperforming state-of-the-art inverse optimal control methods.

Abstract

Assistance games formalize human-robot collaboration under asymmetric information: the human knows the goal, while the robot must infer it from observation and interaction in order to assist effectively. In general, computing optimal assistance game strategies online is intractable, since exact solutions require planning in a POMDP. We identify a class of assistance games in which pragmatic-pedagogic reasoning resolves goal uncertainty in a single time step, rendering the full-horizon game exactly solvable by a tractable best-response procedure. Within this class, we show that mainstream inverse optimal control exhibits an inference ceiling that hinders alignment, while pragmatic-pedagogic reasoning overcomes this barrier by immediately disambiguating goals through actions that look equivalent under task execution alone. Finally, we validate our theoretical results and proposed method on a simple collaborative block-building example.

Problem

• assistance games
• asymmetric information
• goal inference
• human-robot collaboration
• pragmatic-pedagogic reasoning

Innovation

• pragmatic-pedagogic reasoning
• assistance games
• goal disambiguation
• inverse optimal control
• tractable best response