# Researchers report gains from jointly training hand reconstruction and action prediction

_Published Thursday, October 15, 2026 at 11:44 PM EDT · Science, AI, Robotics · Latest · Tier 2 — Notable_

Researchers report that P³-WAM, a model trained jointly to reconstruct hand movements, predict video from actions and plan manipulation sequences, outperformed specialized baselines on hand reconstruction and action-conditioned video prediction across four datasets in a preprint.

The model combines a structured representation of 3D actions with a pretrained video-generation transformer. It can use video and actions as either inputs or prediction targets. Unified training improved all three capabilities over training each mode separately. Planning gains were assessed through a vision-language model, while generated manipulation videos showed stronger instruction adherence and action progression than video-generation baselines.

## Sources

- [arXiv Query: search_query=cat:cs.RO&id_list=&start=0&max_results=30](https://arxiv.org/abs/2610.22363v1)

---
Canonical: https://techandbusiness.org/newswire/cfTOlN4ATICEjfCEM4T5bo
Published: 2026-10-16T03:44:08.762Z
Story chronology: 2026-10-15T17:59:07.000Z
Retrieved: 2026-10-16T06:00:43.138Z
Publisher: Tech & Business (techandbusiness.org)