Science & Tech · AI · 1 hr ago
UBTECH Details Thinker-WM Robot Model
UBTECH has detailed Thinker-WM, a model designed for use with robots.
It brings together video, depth information and robot actions.
A world model is a system intended to represent how the world works to help guide a robot's actions.
UBTECH says Thinker-WM tops three embodied world model benchmarks.
In tests on RoboTwin, adding depth raised single-task success by 15 percentage points.
The results show how depth affected performance in those tests, but no further plans for Thinker-WM were stated.
UBTECH says its Thinker-WM model tops three embodied world model benchmarks.
The model unifies video, depth and robot actions.
UBTECH says adding depth raised single-task success on RoboTwin by 15 percentage points in ablation tests.
- Who
- UBTECH.
- What
- It detailed Thinker-WM, a world model that unifies video, depth and robot actions.
- When
- The article was published on 2026-10-17.
- Where
- Not stated.
- Why
- Not stated.
This story does not have two clearly opposing sides.
This story does not have a timeline yet.
- Model
- Thinker-WM
- Organization
- UBTECH
- Inputs unified
- Video, depth and robot actions
- Benchmarks
- Tops three embodied world model benchmarks, according to UBTECH
- RoboTwin result
- Adding depth lifted single-task success by 15 percentage points in ablation tests