AI Paper Atlas

Research map / 3D vision / 3D generation

GeoMoLa: Geometry-Aware Motion Latents for Learning Robust Manipulation Policies

Yunchao Zhang, Yijia Weng, Ruizhe Liu, Ming Hu, Leonidas Guibas, Yanchao Yang

ICML 2026 · 3D vision › 3D generation · approach: Semantic scene completion

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add pointmap (3D geometry) prediction to motion latent learning (ablation)GeoMoLa · CALVIN ABC->Daverage task chain length: 3.38 -> 3.60Improved
add pointmap (3D geometry) prediction to motion latent learning (ablation)GeoMoLa · RLBench (10 tasks)success rate: 79.6% -> 84.7%Improved
add RGB prediction branch alongside pointmap prediction (ablation)GeoMoLa · CALVIN ABC->Daverage task chain length: 3.50 -> 3.60Improved
add RGB prediction branch alongside pointmap prediction (ablation)GeoMoLa · RLBench (10 tasks)success rate: 82.0% -> 84.7%Improved
add pointmap prediction in real-world training (ablation)GeoMoLa · Real-world manipulationaverage success rate: 35.0% -> 53.3%Improved
learn motion latents from 3D pointmap prediction instead of 2D observations (SkillDiffuser)motion latent policy · RLBench (10 tasks)success rate: 74.4 -> 84.7Improved
use GeoMoLa over 3D Diffuser Actor with limited real demonstrationsreal-world manipulation · Real-world (20 demos per task)average success rate: 40.0% -> 53.3%Improved
removing pointmap prediction on rotation-heavy RLBench tasks (ablation)GeoMoLa · RLBench rotation-heavy taskssuccess rate: nearly 9% dropHurt
removing pointmap prediction on translation-heavy RLBench tasks (ablation)GeoMoLa · RLBench translation-heavy taskssuccess rate: drop less than 1%No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy