arXiv:2605.14211v2 Announce Type: replace-cross 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 unsupervis
The accelerating pace of AI research, particularly in embodied learning and agentic systems, is creating new avenues for self-supervised learning from vast, unstructured internet data.
This development indicates a significant step towards more autonomous and scalable AI learning, potentially overcoming major bottlenecks in data annotation and reward engineering for complex robotic tasks.
AI agents can now learn complex, long-horizon tasks from unlabeled internet video without direct human supervision or reward shaping, accelerating their development and deployment.
- · AI research labs
- · Robotics companies
- · Data infrastructure providers
- · Embodied AI developers
- · Companies reliant on manual data annotation for robotics
- · Developers focused solely on reward-engineered AI
- · Legacy automation providers
- · Companies with limited access to diverse internet video data
More capable and independently learning AI agents will emerge for a wider range of physical and digital tasks.
Reduced dependency on human-curated datasets will lower development costs and accelerate the deployment of autonomous systems in diverse environments.
The proliferation of self-improving, embodied AI agents could fundamentally alter labor markets and the nature of work, especially in fields requiring complex physical interaction.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG