ASH: Agents that Self-Hone via Embodied Learning

• URL: http://arxiv.org/abs/2605.14211v1
• Date: Thu, 14 May 2026 00:10:12 GMT
• Title: ASH: Agents that Self-Hone via Embodied Learning
• Abstract summary: ASH is an agentic system that learns an embodied policy from unlabeled, noisy internet video.<n>ASH uses unsupervised learning to identify key moments from large-scale internet video and retains them as long-term memory.<n>We evaluate ASH on two complementary environments demanding multi-hour planning: Pokemon Emerald, a turn-based RPG, and The Legend of Zelda: The Minish Cap, a real-time action-adventure game.
• Score: 8.529044800084554
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Long-horizon embodied tasks remain a fundamental challenge in AI, as current methods rely on hand-engineered rewards or action-labeled demonstrations, neither of which scales. We introduce ASH, an agentic system that learns an embodied policy from unlabeled, noisy internet video, without reward shaping or expert annotation. ASH follows a self-improvement loop; when it gets stuck, ASH learns an Inverse Dynamics Model (IDM) from its own trajectories, and uses its IDM to extract supervision from relevant internet video. ASH uses unsupervised learning to identify key moments from large-scale internet video and retains them as long-term memory -- allowing it to tackle long-horizon problems. We evaluate ASH on two complementary environments demanding multi-hour planning: Pokemon Emerald, a turn-based RPG, and The Legend of Zelda: The Minish Cap, a real-time action-adventure game. In both games, behavioral cloning, retrieval-augmented and zero-shot foundation-model baselines plateau, while ASH sustains progression across our 8-hour evaluation. ASH reaches an average of $11.2/12$ milestones in Pokemon Emerald and $9.9/12$ in Legend of Zelda, while the strongest baseline gets stuck in both environments at an average of $6.5/12$ and $6.0/12$ milestones, respectively. We demonstrate that self-improving agents are a scalable recipe for long-horizon embodied learning.

Related papers

• PureTD: Reinforcement Learning for Backgammon Money Games with No Evaluation-time Search [51.56484100374058]
  We revisit Tesauro's TDGammon for backgammon money games in the setting of no evaluation-time search.<n>Both checker play and cube action are learned from scratch via self-play reinforcement learning (RL), with minimal hand-coded logic and no expert features.
  arXiv  Detail & Related papers  (2026-08-15T09:46:03Z)
• How to Instruct Your Robot: Dense Language Annotations Power Robot Policy Learning [69.68882580009982]
  We introduce DeMiAn, a two-stage approach that first re-labels demonstration segments with VLM-generated annotations.<n>A learned instructor then maps a task description and initial scene snapshot to a task-appropriate annotation at deployment.<n>On RoboCasa, the instructor raises success by 5 points over a task-only baseline and comes within 3 points of a per-task oracle.
  arXiv  Detail & Related papers  (2026-05-16T16:52:08Z)
• Continual Harness: Online Adaptation for Self-Improving Foundation Agents [21.599574507129365]
  Continual Harness is a reset-free self-improving harness for embodied agents.<n>It drives sustained in-game milestone progress on Pokemon Red without resetting the environment between training iterations.
  arXiv  Detail & Related papers  (2026-05-11T05:21:33Z)
• SAGE: Training Smart Any-Horizon Agents for Long Video Reasoning with Reinforcement Learning [53.67654657011112]
  We propose SAGE, an agent system that performs multi-turn reasoning on long videos while handling simpler problems in a single turn.<n>We further propose an effective RL post-training recipe essential for instilling any-horizon reasoning ability in SAGE-MM.<n>We empirically validate the effectiveness of our system, data, and RL recipe, observing notable improvements of up to 6.1% on open-ended video reasoning tasks.
  arXiv  Detail & Related papers  (2025-12-15T20:14:19Z)
• Game-TARS: Pretrained Foundation Models for Scalable Generalist Multimodal Game Agents [56.25101378553328]
  We present Game-TARS, a generalist game agent trained with a unified, scalable action space anchored to human-aligned keyboard-mouse inputs.<n>Game-TARS is pre-trained on over 500B tokens with diverse trajectories and multimodal data.<n> Experiments show that Game-TARS achieves about 2 times the success rate over the previous sota model on open-world Minecraft tasks.
  arXiv  Detail & Related papers  (2025-10-27T17:43:51Z)
• TimeRewarder: Learning Dense Reward from Passive Videos via Frame-wise Temporal Distance [36.22149703563646]
  TimeRewarder is a simple yet effective reward learning method that derives progress estimation signals from passive videos.<n>We show that TimeRewarder dramatically improves RL for sparse-reward tasks, achieving nearly perfect success in 9/10 tasks with only 200,000 interactions per task with the environment.
  arXiv  Detail & Related papers  (2025-09-30T17:58:20Z)
• FlashAdventure: A Benchmark for GUI Agents Solving Full Story Arcs in Diverse Adventure Games [56.81554611870848]
  We introduce FlashAdventure, a benchmark of 34 Flash-based adventure games designed to test full story arc completion.<n>We also propose CUA-as-a-Judge, an automated gameplay evaluator, and COAST, an agentic framework leveraging long-term clue memory.<n> Experiments show current GUI agents struggle with full story arcs, while COAST improves milestone completion by bridging the observation-behavior gap.
  arXiv  Detail & Related papers  (2025-09-01T01:33:16Z)
• Pokemon Red via Reinforcement Learning [3.548348926427221]
  Pok'emon Red, a classic Game Boy JRPG, presents significant challenges as a testbed for agents.<n>We introduce a simplistic environment and a Deep Reinforcement Learning training methodology, demonstrating a baseline agent that completes an initial segment of the game up to completing Cerulean City.<n>Our experiments include various ablations that reveal vulnerabilities in reward shaping, where agents exploit specific reward signals.
  arXiv  Detail & Related papers  (2025-02-27T09:42:23Z)
• Reward Relabelling for combined Reinforcement and Imitation Learning on sparse-reward tasks [2.0305676256390934]
  We present a new method to leverage demonstrations and episodes collected online in any sparse-reward environment with any off-policy algorithm. Our method is based on a reward bonus given to demonstrations and successful episodes, encouraging expert imitation and self-imitation. Our experiments focus on manipulation robotics, specifically on three tasks for a 6 degrees-of-freedom robotic arm in simulation.
  arXiv  Detail & Related papers  (2022-01-11T08:35:18Z)
• Adversarial Online Learning with Variable Plays in the Pursuit-Evasion Game: Theoretical Foundations and Application in Connected and Automated Vehicle Cybersecurity [5.9774834479750805]
  We extend the adversarial/non-stochastic multi-play multi-armed bandit (MPMAB) to the case where the number of arms to play is variable. The work is motivated by the fact that the resources allocated to scan different critical locations in an interconnected transportation system change dynamically over time and depending on the environment.
  arXiv  Detail & Related papers  (2021-10-26T23:09:42Z)
• Semi-supervised reward learning for offline reinforcement learning [71.6909757718301]
  Training agents usually requires reward functions, but rewards are seldom available in practice and their engineering is challenging and laborious. We propose semi-supervised learning algorithms that learn from limited annotations and incorporate unlabelled data. In our experiments with a simulated robotic arm, we greatly improve upon behavioural cloning and closely approach the performance achieved with ground truth rewards.
  arXiv  Detail & Related papers  (2020-12-12T20:06:15Z)
• Neural MMO v1.3: A Massively Multiagent Game Environment for Training and Evaluating Neural Networks [48.5733173329785]
  We present Neural MMO, a massively multiagent game environment inspired by MMOs. We discuss our progress on two more general challenges in multiagent systems engineering for AI research: distributed infrastructure and game IO.
  arXiv  Detail & Related papers  (2020-01-31T18:50:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.