Moto employs Latent Motion Tokens for unsupervised pretraining, boosting robot manipulation success rate to 61.4%.

Yi Chen, Yuying Ge, Weiliang Tang et al.

G3Flow integrates foundation models with 3D generative and pose tracking to enhance robotic manipulation success rates by over 20% in complex tasks.

Tianxing Chen, Yao Mu, Zhixuan Liang et al.

Proposes Safety-as-Policy combining virtual scenario generation and reflection to enhance robot safety, achieving 52.63% safety rate in synthetic data and 75% in real-world tests.

Minheng Ni, Lei Zhang, Zihan Chen et al.

FoAR combines high-frequency force/torque sensing with vision, using future contact prediction for dynamic multimodal fusion, greatly improving contact-rich manipulation accuracy.

Zihao He, Hongjie Fang, Jingjing Chen et al.

RONAR system leverages large language models to generate natural language descriptions of robot multi-modal experiences, improving failure analysis by 11%.

Zihan Wang, Brian Liang, Varad Dhat et al.

Proposed spatial-aware adversarial patches drastically reduce robot task success, up to 100%, revealing vulnerabilities in VLA models.

Taowen Wang, Cheng Han, James Chenhao Liang et al.

DINO-WM leverages pre-trained visual patch features for zero-shot planning, outperforming state-of-the-art models with 56% LPIPS improvement and 45% success rate increase.

Gaoyue Zhou, Hengkai Pan, Yann LeCun et al.

EgoMimic leverages egocentric human videos and 3D hand tracking with cross-domain alignment and joint training, boosting manipulation task success by 34-228%.

Simar Kareer, Dhruv Patel, Ryan Punamiya et al.

Sparsh uses self-supervised learning on 460k+ tactile images, improving TacBench performance by 95.1%.

Carolina Higuera, Akash Sharma, Chaithanya Krishna Bodduluri et al.

SuctionPrompt integrates VLMs and 3D detection for zero-shot robotic picking, achieving 65% success in diverse environments.

Tomohiro Motoda, Takahide Kitamura, Ryo Hanai et al.

Proposes an adaptive environment-shaping framework using secondary RL (SAC) to enable drone racing in unseen tracks with 100% success rate.

Hongze Wang, Jiaxu Xing, Nico Messikommer et al.

This study reveals that robot imitation learning generalization follows a power-law with environment and object diversity, with 4 hours of data collection achieving ~90% success in new settings.

Fanqi Lin, Yingdong Hu, Pingyue Sheng et al.

Differentiable simulation accelerates quadrotor control training, achieving recovery in seconds with minimal samples, combining simple models and visual features.

Johannes Heeg, Yunlong Song, Davide Scaramuzza

V-GPS leverages offline RL-trained value functions for test-time action re-ranking, boosting five state-of-the-art robotic policies by 82.8% across 12 tasks, with no fine-tuning needed.

Mitsuhiko Nakamoto, Oier Mees, Aviral Kumar et al.

Proposes Gen-Drive, a diffusion-based generative driving policy with reward modeling and RL fine-tuning, achieving state-of-the-art planning scores on nuPlan.

Zhiyu Huang, Xinshuo Weng, Maximilian Igl et al.

FeelAnyForce uses a multi-head Transformer with RGB and depth images to estimate 3D contact forces, achieving 4% MAE on unseen objects.

Amir-Hossein Shahidzadeh, Gabriele Caddeo, Koushik Alapati et al.

ManiSkill3 is a GPU-parallelized robotics simulator supporting 12 task domains, achieving 30,000+ FPS with 2-3x less memory, enabling fast visual RL training.

Stone Tao, Fanbo Xiang, Arth Shukla et al.

COGIS employs Gaussian Process Implicit Surfaces to online model obstacles, enhancing robot manipulation in partial observability with constraint-based dataset refinement.

Abhinav Kumar, Peter Mitrano, Dmitry Berenson

Human-centered quadruped guide dog robot utilizing visual foundation models and RL control, achieving 95% obstacle avoidance success and 2-hour battery life.

Hochul Hwang, Ken Suzuki, Nicholas A Giudice et al.

RoboNurse-VLA integrates SAM 2 and Llama 2 for high-precision surgical instrument handover, achieving over 90% success in unseen scenarios.

Shunlei Li, Jin Wang, Rui Dai et al.