AI Paper Atlas

Research map / Vision-language and multimodal / Robot action and navigation

Learning to Grasp Anything By Playing with Random Toys

Dantong Niu, Yuvan Sharma, Baifeng Shi, Rachel Ding, Matteo Gioia, Haoru Xue, Henry Tsai, Konstantinos Kallidromitis, Anirudh Pai, S. Shankar Sastry, Trevor Darrell, Jitendra Malik and 1 more

ICLR 2026 · Vision-language and multimodal › Robot action and navigation · approach: Memory and subgoal-based manipulation · 1 citations (OpenAlex)

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
use detection pooling instead of attention pooling (ablation)LEGO · ManiSkill zero-shot grasping on YCB (65 objects), 2500 demossuccess rate (%): 51.63 -> 80.00Improved
use detection pooling instead of CLS pooling (ablation)LEGO · ManiSkill zero-shot grasping on YCB (65 objects), 2500 demossuccess rate (%): 49.81 -> 80.00Improved
use detection pooling instead of mean pooling (ablation)LEGO · ManiSkill zero-shot grasping on YCB (65 objects), 2500 demossuccess rate (%): 40.58 -> 80.00Improved
use absolute joint control instead of delta control (ablation)LEGO · ManiSkill zero-shot grasping on YCB (65 objects), 2500 demossuccess rate (%): 37.02 -> 80.00Improved
train on toys with four colors instead of red only (ablation)LEGO · ManiSkill zero-shot grasping on YCB (65 objects), 2500 demossuccess rate (%): 76.35 -> 80Improved
include sphere primitives in toys (ablation)LEGO · ManiSkill grasping, primitive removed from toys, 1000 demossuccess rate (%): 63.08 -> 72.12Improved
include ring primitives in toys (ablation)LEGO · ManiSkill grasping, primitive removed from toys, 200 demossuccess rate (%): 72.6 -> 72.12No clear effect
train on two-primitive toys rather than five-primitive toys (ablation)LEGO · ManiSkill grasping, toy complexity (250 demos)success rate (%): 10.19 -> 44.42Improved
increase condition length c from 4 to 16 (ablation)LEGO · ManiSkill zero-shot grasping on YCB (65 objects), 2500 demossuccess rate (%): 54.30 -> 80.00Improved
increase prediction length k from 4 to 16 (ablation)LEGO · ManiSkill zero-shot grasping on YCB (65 objects), 2500 demossuccess rate (%): 70.67 -> 80.00Improved
increase prediction length k from 4 to 16 (ablation)LEGO · ManiSkill zero-shot grasping on YCB (65 objects), 250 demossuccess rate (%): 63.65 -> 56.63Hurt
scale policy transformer from ViT-Base (86M) to ViT-Large (307M) (ablation)LEGO · ManiSkill zero-shot grasping on YCB (65 objects)success rate (%)No clear effect

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