Robotics

Dita: Scales a diffusion transformer for generalist robot policies, enabling 10-shot learning in complex, real-world tasks.

API vs. GUI Agents: Understanding the divergence and convergence in LLM-based automation.

ADC: Human-robot collaboration revolutionizes data collection, slashing data needs and boosting robot learning!

KUDA unifies dynamics learning and visual prompting with keypoints for open-vocabulary robot manipulation.

BRS: Streamlining real-world whole-body manipulation for household activities. It introduces a robot suite tackling robot dexterity with bimanual coordination, navigation, and end-effector reach.

CognitiveDrone: A novel VLA model and benchmark for real-time cognitive UAV tasks, improving reasoning and control.

Sim-to-real RL recipe achieves robust vision-based dexterous humanoid manipulation without human demos!

Reflect VLM: Improving robotic manipulation via vision-language models with a novel reflection mechanism and a diffusion model for imagined futures.

ARM4R pre-trains autoregressive robotic models using low-level 4D representations from human videos, achieving efficient transfer learning and improved task performance across various environments.

HUMANUP: A novel two-stage reinforcement learning framework enables real-world humanoid robots to autonomously recover from falls on various terrains.

DexTrack achieves highly generalizable neural tracking control for dexterous robot manipulation by iteratively training a controller using high-quality demonstrations refined via homotopy optimization…

HMA: a novel approach for generating high-quality robotic videos 15x faster, enabling real-time policy evaluation and data augmentation for scaling robot learning.

FAST: A novel action tokenization method using discrete cosine transform drastically improves autoregressive vision-language-action models’ training and performance, enabling dexterous and high-freque…

EnerVerse: A novel framework seamlessly integrates convolutional and attention mechanisms to generate embodied future spaces for enhanced robotic manipulation, mitigating data scarcity with a generati…

SWE-Gym, a novel environment for training real-world software engineering agents using 2,438 real-world Python task instances, achieves new state-of-the-art performance and is publicly available.

MoDE makes AI for robot control faster and more efficient.

From language models to action models: building AI that does things.

TidyBot++: Low-cost, open-source holonomic mobile base makes robot learning easier.

CARP: A novel visuomotor policy learning paradigm achieves high accuracy and 10x faster inference than state-of-the-art by combining autoregressive efficiency and diffusion model precision through a c…

RAPL efficiently aligns robots with human preferences using minimal feedback by aligning visual representations before reward learning.