# Robotic Task Ambiguity Resolution via Natural Language Interaction

**Authors:** Eugenio Chisari, Jan Ole von Hartz, Fabien Despinoy, Abhinav Valada  
**arXiv:** [2504.17748](https://arxiv.org/abs/2504.17748)  
**Date:** 2025-04-24  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2504.17748v1/

## Key points

- Language-conditioned policies in robotics enable users to articulate tasks using natural language, enhancing versatility.
- Ambiguities in task descriptions can lead to failures in robotic agents' interpretation and execution of tasks.
- AmbResVLM is a groundbreaking method that grounds language goals in the observed scene and addresses task ambiguity through reasoning.
- Extensive evaluations show that AmbResVLM effectively detects and resolves task ambiguities, leading to a significant boost in downstream robot policy performance.
- Researchers have openly shared data, code, and trained models for AmbResVLM at https://ambres.cs.uni-freiburg.de.

## Summary

, , , , 

In the realm of robotics, language-conditioned policies have become increasingly popular as they enable users to articulate tasks using natural language, thereby enhancing their versatility. Significant research efforts have been dedicated to improving the action prediction capabilities of these policies. However, there has been a notable oversight in addressing the challenge of reasoning about task descriptions. Ambiguities in task descriptions often result in failures of downstream policies due to misinterpretation by robotic agents. To tackle this issue, a groundbreaking method called AmbResVLM has been introduced. This innovative approach grounds language goals in the observed scene and explicitly addresses task ambiguity through reasoning. Through extensive evaluations conducted across simulated and real-world domains, AmbResVLM has demonstrated remarkable effectiveness in detecting and resolving task ambiguities when compared to recent state-of-the-art baselines. Real robot experiments further validate the efficacy of AmbResVLM by showcasing a substantial improvement in the performance of downstream robot policies. The average success rate has seen a significant boost from 69.6% to an impressive 97.1%. Moreover, the researchers behind this method have made their data, code, and trained models openly accessible at https://ambres.cs.uni-freiburg.de. Authored by Eugenio Chisari, Jan Ole von Hartz, Fabien Despinoy, and Abhinav Valada, the study titled "Robotic Task Ambiguity Resolution via Natural Language Interaction" sheds light on the critical importance of addressing task ambiguity in robotics through innovative methods like AmbResVLM.