Xueyuan Wei, Xuanxu Wang, Xudong Li et al.
Fine-grained action annotations can support skill reuse and compositional generalization in robot policies.Fine-grained action annotations can support skill reuse and compositional generalization in robot policies. However, robot datasets typically provide task-level instructions without subtask annotations.
Ian Chuang, Hiya Shah, Vamsi Chilamkur et al.
Dexterous manipulation with multi-fingered hands is the next frontier of robotics, yet learned hand skills remain limited to simple grasps or basic repeated motions, such…Dexterous manipulation with multi-fingered hands is the next frontier of robotics, yet learned hand skills remain limited to simple grasps or basic repeated motions, such as rotating an object.
Dequan Ou, Austin Lopez, Quentin Missinne et al.
Event cameras offer low latency, high dynamic range, and milliwatt-level power consumption, making them ideal for agile robots such as tiny drones navigating through…Event cameras offer low latency, high dynamic range, and milliwatt-level power consumption, making them ideal for agile robots such as tiny drones navigating through cluttered environments.
Mert Gürbüzbalaban, M. Hamza Rafi, Necdet Serhat Aybat et al.
With exact, delay-free gradients, Nesterov's accelerated gradient (NAG) achieves optimal √κ dependence for L-smooth, μ-strongly convex objectives, where κ=L/μ; heavy-ball…With exact, delay-free gradients, Nesterov's accelerated gradient (NAG) achieves optimal √κ dependence for L-smooth, μ-strongly convex objectives, where κ=L/μ; heavy-ball (HB) achieves the same scaling on strongly convex…
Buzhao Liu, Jianing Ye, Xiu Yuan et al.
Recently, LLM-based autonomous agents have shown strong capabilities in iteratively generating, evaluating, and refining executable solutions.Recently, LLM-based autonomous agents have shown strong capabilities in iteratively generating, evaluating, and refining executable solutions.
Borui Xu, Hongshi Tan, Wenqi Jiang et al.
Vision-language-action (VLA) models increasingly run on edge devices aboard robots, where inference latency constrains closed-loop robot control.Vision-language-action (VLA) models increasingly run on edge devices aboard robots, where inference latency constrains closed-loop robot control.
Hongcheng Gao, Zelin Zheng, Alex Zhou et al.
Embodied agents are expected to accomplish real-world goals by deciding for themselves how and when to interact with their surroundings.Embodied agents are expected to accomplish real-world goals by deciding for themselves how and when to interact with their surroundings.
Haoyu Zhao, Xingyue Zhao, Zhian Su et al.
Scaling vision-language-action (VLA) models is constrained by scarce action-labeled demonstrations and incompatible control spaces across robot embodiments.Scaling vision-language-action (VLA) models is constrained by scarce action-labeled demonstrations and incompatible control spaces across robot embodiments.
Lexiao Lai
We construct a coercive, locally Lipschitz, semialgebraic function on mathbb R² for which BFGS with Armijo–Wolfe line search converges to a non-Clarke-stationary point…We construct a coercive, locally Lipschitz, semialgebraic function on mathbb R² for which BFGS with Armijo–Wolfe line search converges to a non-Clarke-stationary point from a nonempty open set of initial points and inverse-Hessian…
Doanh Le Thien, Viet-Thanh Nguyen, Pham Tri Quang et al.
Video world models (VWMs) are rapidly emerging as scalable neural simulators for robotic manipulation. However, most benchmarks only evaluate them as open-world video…Video world models (VWMs) are rapidly emerging as scalable neural simulators for robotic manipulation. However, most benchmarks only evaluate them as open-world video generators, using Vision-Language Models (VLMs) that lack…
Probes showed at least 79% of updates were erased, and adding a 32-bit copy raised task success from 89.0% to 97.7% for π_(0.5).Probes showed at least 79% of updates were erased, and adding a 32-bit copy raised task success from 89.0% to 97.7% for π_(0.5). This empirical study examined fine-tuning recipes for vision-language-action models, which translate…
Jiawei Chen, Yinxi Wang, Zihan Yang et al.
What visual representation should a navigation policy act on? Navigation policies built on vision foundation models largely leave representation design to the encoders'…What visual representation should a navigation policy act on? Navigation policies built on vision foundation models largely leave representation design to the encoders' pretraining objectives.
Mohsen Amiri, Sima Didari, Davoud Ataee Tarzanagh et al.
Motion planning is central to robotics, and it is increasingly used to correct the motions proposed by learned planners and policies.Motion planning is central to robotics, and it is increasingly used to correct the motions proposed by learned planners and policies. In both uses, constraints must hold at every instant of the motion, not only at sampled times.
Xinxin Liu, Bingxin Xu, Wenrui Bao et al.
Agentic robotics systems combine planners with robot harnesses to perform complex manipulation tasks. Recent work improves these systems through better harnesses and…Agentic robotics systems combine planners with robot harnesses to perform complex manipulation tasks. Recent work improves these systems through better harnesses and external memory, but relies on powerful closed-source models as…
Christopher Leet, Achyutha Menon, Sravanthi Machcha et al.
General purpose multimodal language models (LMs) are increasingly capable of directly controlling robotic arms (e.g., by commanding end effector poses) to manipulate…General purpose multimodal language models (LMs) are increasingly capable of directly controlling robotic arms (e.g., by commanding end effector poses) to manipulate their environment.
Sriyash Poddar, Yanda Bao, Jacob Krantz et al.
Sequential decision-making under partial observability is difficult because agents must explore to infer hidden, task-relevant information before acting effectively.Sequential decision-making under partial observability is difficult because agents must explore to infer hidden, task-relevant information before acting effectively.
Mingze Sun, Ziang Cao, Yuan Fei et al.
Despite recent progress in dexterous reference tracking, existing models focus primarily on single-hand manipulation and a limited range of interactions, leaving the…Despite recent progress in dexterous reference tracking, existing models focus primarily on single-hand manipulation and a limited range of interactions, leaving the bimanual dexterous behaviors common in real life unaddressed.
Rangya Zhang, Yujing Shang, Zhuoyuan Yu et al.
Large-scale aerial resources provide a foundation for learning navigation across diverse environments. However, existing resources remain limited in providing (1) densely…Large-scale aerial resources provide a foundation for learning navigation across diverse environments. However, existing resources remain limited in providing (1) densely sampled flight demonstrations for learning different policy…
Qiayuan Liao, Yiyang Shao, Junfeng Long et al.
Tracking large human-motion datasets has become the dominant route to general, expressive humanoid whole-body control, but scaling across many motions is expensive.Tracking large human-motion datasets has become the dominant route to general, expressive humanoid whole-body control, but scaling across many motions is expensive.
Hongcheng Gao, Alex Zhou, Jay Zhu et al.
Agents built on vision language models (VLMs) now compete with learned policies on robot manipulation leaderboards.Agents built on vision language models (VLMs) now compete with learned policies on robot manipulation leaderboards. We ask how far a coding agent can go when it controls a robot instead of a computer, and where it stops.
Emad Aghajanzadeh, Thomas Schwarzfischer, Moritz Thoma et al.
Closed-loop evaluation is widely regarded as a more faithful way to assess end-to-end (E2E) driving models than open-loop evaluation, and Bench2Drive, built on the CARLA…Closed-loop evaluation is widely regarded as a more faithful way to assess end-to-end (E2E) driving models than open-loop evaluation, and Bench2Drive, built on the CARLA simulator, has become the dominant benchmark for it.
Tianwei Luo, Yiming He, Hongjin Wu et al.
General-purpose manipulation requires reliable behavior across embodiments, objects, and task demands. We introduce HandsOn, a composable simulation platform and…General-purpose manipulation requires reliable behavior across embodiments, objects, and task demands. We introduce HandsOn, a composable simulation platform and benchmark for cross-embodiment dexterous manipulation.
Chenglong Bao, Liang Chen, Jiahong Li et al.
Gradient restarting often improves the practical performance of accelerated gradient methods for unconstrained convex optimization problems, but a general theoretical…Gradient restarting often improves the practical performance of accelerated gradient methods for unconstrained convex optimization problems, but a general theoretical guarantee of such improvement remains unclear.
Xueyang Zhou, Yangming Xu, Qianjiang Li et al.
LIBERO has emerged as a widely adopted benchmark for evaluating robotic manipulation policies. However, strong performance on LIBERO does not necessarily indicate robust…LIBERO has emerged as a widely adopted benchmark for evaluating robotic manipulation policies. However, strong performance on LIBERO does not necessarily indicate robust embodied capabilities, as models may exploit…
Antonia Bronars, Younghyo Park, Kushal Kedia et al.
Every robot already carries high-resolution contact sensors: its own joints. Legged robots and humanoids exploit them to walk, run, and track whole-body motions without…Every robot already carries high-resolution contact sensors: its own joints. Legged robots and humanoids exploit them to walk, run, and track whole-body motions without vision, but there the contacts are large and forceful.
Lingyi Hong, Weiren Cai, Jiayu Chen et al.
Vision-and-Language Navigation (VLN) requires embodied agents to ground language in visual observations and make long-horizon decisions.Vision-and-Language Navigation (VLN) requires embodied agents to ground language in visual observations and make long-horizon decisions. Existing methods often address these demands through task-specific architectures.
Runhan Huang, Xiangcheng Zhang, Haoyu Chen et al.
Robot learning has advanced by scaling demonstration datasets and model capacity; however, deploying a new robot embodiment or task typically necessitates a dedicated…Robot learning has advanced by scaling demonstration datasets and model capacity; however, deploying a new robot embodiment or task typically necessitates a dedicated teleoperation corpus.
Jack B. Jedlicki, Amish Sethi, Heng Yang
We use a frozen frontier vision-language model, GPT-6-Astra, directly as the robot policy on RoboMME, a benchmark of 16 memory-dependent manipulation tasks spanning…We use a frozen frontier vision-language model, GPT-6-Astra, directly as the robot policy on RoboMME, a benchmark of 16 memory-dependent manipulation tasks spanning temporal, spatial, object, and procedural memory.
Fucai Zhu, Xidong Zhang, Yichi Zhang et al.
Large-scale robot learning increasingly relies on diverse human demonstrations, yet tactile manipulation data remain difficult to scale, and existing multimodal datasets…Large-scale robot learning increasingly relies on diverse human demonstrations, yet tactile manipulation data remain difficult to scale, and existing multimodal datasets often lack consistent quality control and structured…
Hanlong Li, Sihan Tan, Takeshi Ashizawa et al.
Vision-language-action (VLA) policies fine-tuned on LIBERO are trained and evaluated on the same ten instruction strings, and rewording the instruction changes base…Vision-language-action (VLA) policies fine-tuned on LIBERO are trained and evaluated on the same ten instruction strings, and rewording the instruction changes base OpenVLA's success by -38.0 points.
Andrea Yaoyun Cui, Hanyu Wang, Xinyi Wang et al.
Robot policies are expected to generalize across deployment environments, including variations in the robot itself: cameras are bumped and remounted, calibrations drift…Robot policies are expected to generalize across deployment environments, including variations in the robot itself: cameras are bumped and remounted, calibrations drift, controllers run at different rates, and end-effector pads…
Xinyi Wang, Bangzheng Li, Hanyu Wang et al.
Frontier language-model agents can now improve robots on their own: they write control code, test it, and keep what works as skills and memories for the next task.Frontier language-model agents can now improve robots on their own: they write control code, test it, and keep what works as skills and memories for the next task.
Gennady Sidorov, Valerii Kornilov, Denis Grankin et al.
A robot needs a world model that predicts how its actions will change the scene and a policy that chooses those actions.A robot needs a world model that predicts how its actions will change the scene and a policy that chooses those actions. Models are usually built and evaluated for only one of these roles, and every robot records its actions in…
Zihan You, Zehua Fan, Wenqi Lyu et al.
Recent Vision-Language-Action (VLA) models increasingly use natural-language instructions to condition autonomous driving behavior.Recent Vision-Language-Action (VLA) models increasingly use natural-language instructions to condition autonomous driving behavior.
On a navigation benchmark, NavGPT-3 scored 90.43 in success against 90.4 for human followers, and 78.47 in path fidelity against 77.7.On a navigation benchmark, NavGPT-3 scored 90.43 in success against 90.4 for human followers, and 78.47 in path fidelity against 77.7. In a preprint, researchers built NavGPT-3, a harness and runtime system that connects a…
Gon Buzaglo, Guy Kornowski
We present new algorithms for stochastic optimization problems in which the objective is neither smooth nor convex, subject to a convex constraint set.We present new algorithms for stochastic optimization problems in which the objective is neither smooth nor convex, subject to a convex constraint set.
Justin Tien-Smith, Matthew Pidden, Rolandos Alexandros Potamias
Vision-language-action models (VLAs) offer a promising route toward general-purpose robotics by transferring semantic knowledge from large-scale vision-language…Vision-language-action models (VLAs) offer a promising route toward general-purpose robotics by transferring semantic knowledge from large-scale vision-language pretraining into language-conditioned control.
Octi Zhang, Mateo Guaman Castro, Patrick Yin et al.
General-purpose robots must perform a wide range of tasks from agile locomotion to dexterous manipulation.General-purpose robots must perform a wide range of tasks from agile locomotion to dexterous manipulation. While sim-to-real reinforcement learning (RL) has proven to be a useful tool for this goal, current RL pipelines depend on…
Mert Albaba, Jens Beißwenger, Anna Manasyan et al.
Teaching a humanoid to follow instructions with its whole body runs into two obstacles. Its action space is large and tightly coupled: legs, arms, and fingers must move…Teaching a humanoid to follow instructions with its whole body runs into two obstacles. Its action space is large and tightly coupled: legs, arms, and fingers must move together while the robot keeps its balance, which makes…
Gokul Puthumanaillam, Tao Sun, Elie Aljalbout et al.
The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale.The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale.