From Failures to Learnable Worlds: Policy-Conditioned Environment Generation for Navigation Agents

Hugging Face Daily Papers Papers
embodied-navigation environment-generation llm-agent reinforcement-learning curriculum-learning failure-diagnosis

Summary

WEDGE 是一个弱点引导的导航学习框架，利用 LLM 从预训练策略的成功与失败轨迹中诊断失败模式，并据此生成针对性训练环境，在 ProcTHOR 和 HM3D 上显著优于随机与难例挖掘基线。

Simulated environments enable scalable training data generation for embodied agents, but more data does not necessarily provide what a particular agent needs to learn. Existing environment-generation pipelines largely optimize generic properties such as realism, diversity, or task coverage, while different pretrained policies can fail for different environmental reasons. We therefore introduce WEDGE, a framework for weakness-guided navigation learning in which the failure modes of a pretrained agent guide the generation of training environments at each adaptation round. Given a pretrained policy, WEDGE uses an LLM to propose executable weakness features from contrasting successful and failed rollouts, automatically validates them on the diagnostic set, and selects the strongest validated weakness based on its success-rate gap. The validated weaknesses are then mapped to the environmental factors that control them. WEDGE varies these factors to construct environments in the difficulty regimes where the policy struggles, producing targeted training experience for adaptation. Across three navigation agents with different adaptation mechanisms, we find that policies exhibit distinct weakness profiles and benefit from different training environments. Weakness-conditioned adaptation consistently outperforms matched random and hard-example alternatives on ProcTHOR, while the resulting gains for Uni-NaVid and Gemini also extend to HM3D after ProcTHOR adaptation. Iterative re-diagnosis further shows that dominant weaknesses can persist or shift as the policy improves, providing updated guidance for subsequent environment construction. Together, these results show that policy failures are not only diagnostic signals, but actionable guidance for constructing training environments that target a policy's current weaknesses.
Original Article
View Cached Full Text

Cached at: 10/15/26, 02:58 PM

Paper page - From Failures to Learnable Worlds: Policy-Conditioned Environment Generation for Navigation Agents

Source: https://huggingface.co/papers/2610.19420

Abstract

WEDGE guides environment generation for embodied navigation agents by diagnosing policy failure modes and constructing targeted training scenarios that address specific weaknesses.

Simulated environments enable scalable training data generation for embodied agents, but more data does not necessarily provide what a particular agent needs to learn. Existing environment-generation pipelines largely optimize generic properties such as realism, diversity, or task coverage, while different pretrained policies can fail for different environmental reasons. We therefore introduceWEDGE, a framework forweakness-guided navigation learningin which thefailure modesof a pretrained agent guide the generation of training environments at each adaptation round. Given a pretrained policy,WEDGEuses anLLMto propose executableweakness featuresfrom contrasting successful and failed rollouts, automatically validates them on thediagnostic set, and selects the strongest validated weakness based on itssuccess-rate gap. The validated weaknesses are then mapped to theenvironmental factorsthat control them.WEDGEvaries these factors to construct environments in thedifficulty regimeswhere the policy struggles, producing targeted training experience for adaptation. Across three navigation agents with different adaptation mechanisms, we find that policies exhibit distinct weakness profiles and benefit from different training environments. Weakness-conditioned adaptation consistently outperforms matched random and hard-example alternatives onProcTHOR, while the resulting gains forUni-NaVidandGeminialso extend toHM3DafterProcTHORadaptation.Iterative re-diagnosisfurther shows that dominant weaknesses can persist or shift as the policy improves, providing updated guidance for subsequent environment construction. Together, these results show that policy failures are not only diagnostic signals, but actionable guidance for constructing training environments that target a policy’s current weaknesses.

View arXiv pageView PDFProject pageGitHubAdd to collection

Get this paper in your agent:

hf papers read 2610\.19420

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.19420 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.19420 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.19420 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

From Trainee to Trainer: LLM-Designed Training Environment for RL with Multi-Agent Reasoning

arXiv cs.CL

This paper proposes the LLM-as-Environment-Engineer framework, where a policy model analyzes failures to automatically redesign the training environment for reinforcement learning, and introduces MAPF-FrozenLake as a controllable testbed. The framework, using Qwen3-4B, outperforms larger models like GPT and Gemini, showing that policy learning improves the model's ability to diagnose weaknesses.

Beyond the Remembered World: Predictive 4D Belief for Persistent Navigation in Evolving Worlds

Hugging Face Daily Papers

The paper introduces EvolvingNav, a predictive 4D belief framework for embodied agents that tracks targets in evolving environments, forecasts their locations at inspection time, and revises beliefs under limited visibility, along with the EvoWorld-Bench benchmark covering 54 scenes and 803,680 tasks, demonstrating improved navigation success in simulation and real-robot experiments.

Learning Safe Agent Behaviour from Human Preferences and Justifications via World Models

arXiv cs.AI

This paper introduces DROPJ, a human-centred method for safely training and deploying agent policies by learning a world model from real-world trajectories, then eliciting human preferences with justifications to train a reward model for model predictive control. Experiments show that using human-generated simulated trajectories and justifications improves safety and reduces computational cost.