One of the most important, yet challenging, skills for a dexterous robot is grasping a diverse range of objects. Much of the prior work has been limited by speed, generality, or reliance on depth maps and object poses. In this paper, we introduce DextrAH-RGB, a system that can perform dexterous arm-hand grasping end-to-end from RGB image input. We train a policy in simulation through reinforcement learning that acts on a geometric fabric controller to dexterously grasp a wide variety of objects. We then distill this into an RGB-based policy strictly in simulation using photorealistic tiled rendering. To our knowledge, this is the first work that is able to demonstrate robust sim-to-real transfer of an end-to-end (monocular or stereo) RGB-based policy for complex, dynamic, contact-rich tasks such as dexterous grasping with multi-fingered hands. Unlike previous methods, DextrAH-RGB requires no explicit depth or CAD models, making it significantly more practical and robust in varied real-world lighting and texture conditions. It generalizes to novel objects and scenes, offering a strong step toward deployable, vision-based dexterous manipulation.

We introduce Accelerating Dexterity via Pre-Training (ADEPT), a large-scale reinforcement learning (RL) framework for learning sim-to-real transferable dexterity across high degree-of-freedom (DoF) robot embodiments that can solve long-horizon tasks directly from raw visuo-tactile perception. ADEPT pretrains a dexterous policy on a generic object reposing task, then post-trains downstream policies with this pretrained behavior as a prior. ADEPT enables learning new behaviors that are otherwise difficult to discover from scratch on multi-fingered robots and avoids learning the same set of skills over again for every new downstream task. The pretrained policy zero-shots the reposing phase of downstream tasks, but naïve RL fine-tuning rapidly degrades this capability during transfer. We address this with a stable post-training recipe combining behavior-cloning distillation, critic warm-up, and conservative on-policy updates. To safely exploit the full kinematic dexterity, we introduce a joint-space Geometric Fabric that mediates between the RL policy and the robot. We distill post-trained teachers into perceptive students that zero-shot sim-to-real transfer on two embodiments: a 23 DoF Kuka-Allegro with two RGB cameras, and a 29 DoF Flexiv-Sharpa with two RGB cameras and five vision-based tactile sensors, and can solve long-horizon tasks from challenging initial states with dexterity at human-level speed.

Autonomous multi-finger grasping is a fundamental capability in robotic manipulation. Optimization-based approaches show strong performance, but tend to be sensitive to initialization and are potentially time-consuming. As an alternative, the generator-evaluator-planner framework has been proposed. A generator generates grasp candidates, an evaluator ranks the proposed grasps, and a motion planner plans a trajectory to the highest-ranked grasp. If the planner doesn't find a trajectory, a new trajectory optimization is started with the next-best grasp as the target and so on. However, executing lower-ranked grasps means a lower chance of grasp success, and multiple trajectory optimizations are time-consuming. Alternatively, relaxing the threshold for motion planning accuracy allows for easier computation of a successful trajectory but implies lower accuracy in estimating grasp success likelihood. It's a lose-lose proposition: either spend more time finding a successful trajectory or have a worse estimate of grasp success. We propose a framework that plans trajectories to a set of generated grasp targets, the evaluator estimates the grasp success likelihood at the terminal configuration of the planned trajectories, and the robot executes the trajectory most likely to succeed. Our experiments show our approach improves over the traditional generator-evaluator-planner framework across different objects, generators, and motion planners, and successfully generalizes to novel environments in the real world, including different shelves and table heights. Project Website: https://martinmatak.github.io/fpte/

We present Isaac Lab, the natural successor to Isaac Gym, which extends the paradigm of GPU-native robotics simulation into the era of large-scale multi-modal learning. Isaac Lab combines high-fidelity GPU parallel physics, photorealistic rendering, and a modular, composable architecture for designing environments and training robot policies. Beyond physics and rendering, the framework integrates actuator models, multi-frequency sensor simulation, data collection pipelines, and domain randomization tools, unifying best practices for reinforcement and imitation learning at scale within a single extensible platform. We highlight its application to a diverse set of challenges, including whole-body control, cross-embodiment mobility, contact-rich and dexterous manipulation, and the integration of human demonstrations for skill acquisition. Finally, we discuss upcoming integration with the differentiable, GPU-accelerated Newton physics engine, which promises new opportunities for scalable, data-efficient, and gradient-based approaches to robot learning. We believe Isaac Lab's combination of advanced simulation capabilities, rich sensing, and data-center scale execution will help unlock the next generation of breakthroughs in robotics research.

We present Isaac Lab, the natural successor to Isaac Gym, which extends the paradigm of GPU-native robotics simulation into the era of large-scale multi-modal learning. Isaac Lab combines high-fidelity GPU parallel physics, photorealistic rendering, and a modular, composable architecture for designing environments and training robot policies. Beyond physics and rendering, the framework integrates actuator models, multi-frequency sensor simulation, data collection pipelines, and domain randomization tools, unifying best practices for reinforcement and imitation learning at scale within a single extensible platform. We highlight its application to a diverse set of challenges, including whole-body control, cross-embodiment mobility, contact-rich and dexterous manipulation, and the integration of human demonstrations for skill acquisition. Finally, we discuss upcoming integration with the differentiable, GPU-accelerated Newton physics engine, which promises new opportunities for scalable, data-efficient, and gradient-based approaches to robot learning. We believe Isaac Lab's combination of advanced simulation capabilities, rich sensing, and data-center scale execution will help unlock the next generation of breakthroughs in robotics research.

This work explores techniques to scale up image-based end-to-end learning for dexterous grasping with an arm + hand system. Unlike state-based RL, vision-based RL is much more memory inefficient, resulting in relatively low batch sizes, which is not amenable for algorithms like PPO. Nevertheless, it is still an attractive method as unlike the more commonly used techniques which distill state-based policies into vision networks, end-to-end RL can allow for emergent active vision behaviors. We identify a key bottleneck in training these policies is the way most existing simulators scale to multiple GPUs using traditional data parallelism techniques. We propose a new method where we disaggregate the simulator and RL (both training and experience buffers) onto separate GPUs. On a node with four GPUs, we have the simulator running on three of them, and PPO running on the fourth. We are able to show that with the same number of GPUs, we can double the number of existing environments compared to the previous baseline of standard data parallelism. This allows us to train vision-based environments, end-to-end with depth, which were previously performing far worse with the baseline. We train and distill both depth and state-based policies into stereo RGB networks and show that depth distillation leads to better results, both in simulation and reality. This improvement is likely due to the observability gap between state and vision policies which does not exist when distilling depth policies into stereo RGB. We further show that the increased batch size brought about by disaggregated simulation also improves real world performance. When deploying in the real world, we improve upon the previous state-of-the-art vision-based results using our end-to-end policies.

Vision-based object detectors are a crucial basis for robotics applications as they provide valuable information about object localization in the environment. These need to ensure high reliability in different lighting conditions, occlusions, and visual artifacts, all while running in real-time. Collecting and annotating real-world data for these networks is prohibitively time consuming and costly, especially for custom assets, such as industrial objects, making it untenable for generalization to in-the-wild scenarios. To this end, we present Synthetica, a method for large-scale synthetic data generation for training robust state estimators. This paper focuses on the task of object detection, an important problem which can serve as the front-end for most state estimation problems, such as pose estimation. Leveraging data from a photorealistic ray-tracing renderer, we scale up data generation, generating 2.7 million images, to train highly accurate real-time detection transformers. We present a collection of rendering randomization and training-time data augmentation techniques conducive to robust sim-to-real performance for vision tasks. We demonstrate state-of-the-art performance on the task of object detection while having detectors that run at 50-100Hz which is 9 times faster than the prior state-of-the-art (SOTA). We further demonstrate the usefulness of our training methodology for robotics applications by showcasing a pipeline for use in the real world with custom objects for which there do not exist prior datasets. Our work highlights the importance of scaling synthetic data generation for robust sim-to-real transfer while achieving the fastest real-time inference speeds. Videos and supplementary information can be found at https://sites.google.com/ view/synthetica-vision

A pivotal challenge in robotics is achieving fast, safe, and robust dexterous grasping across a diverse range of objects, an important goal within industrial applications. However, existing methods often have very limited speed, dexterity, and generality, along with limited or no hardware safety guarantees. In this work, we introduce DextrAH-G, a depth-based dexterous grasping policy trained entirely in simulation that combines reinforcement learning, geometric fabrics, and teacher-student distillation. We address key challenges in joint arm-hand policy learning, such as high-dimensional observation and action spaces, the sim2real gap, collision avoidance, and hardware constraints. DextrAH-G enables a 23 motor arm-hand robot to safely and continuously grasp and transport a large variety of objects at high speed using multi-modal inputs including depth images, allowing generalization across object geometry. Videos at https://sites.google.com/view/dextrah-g.

Robotics policies are always subjected to complex, second order dynamics that entangle their actions with resulting states. In reinforcement learning (RL) contexts, policies have the burden of deciphering these complicated interactions over massive amounts of experience and complex reward functions to learn how to accomplish tasks. Moreover, policies typically issue actions directly to controllers like Operational Space Control (OSC) or joint PD control, which induces straightline motion towards these action targets in task or joint space. However, straightline motion in these spaces for the most part do not capture the rich, nonlinear behavior our robots need to exhibit, shifting the burden of discovering these behaviors more completely to the agent. Unlike these simpler controllers, geometric fabrics capture a much richer and desirable set of behaviors via artificial, second order dynamics grounded in nonlinear geometry. These artificial dynamics shift the uncontrolled dynamics of a robot via an appropriate control law to form behavioral dynamics. Behavioral dynamics unlock a new action space and safe, guiding behavior over which RL policies are trained. Behavioral dynamics enable bang-bang-like RL policy actions that are still safe for real robots, simplify reward engineering, and help sequence real-world, high-performance policies. We describe the framework more generally and create a specific instantiation for the problem of dexterous, in-hand reorientation of a cube by a highly actuated robot hand.

Coordinating the motion of robots with high degrees of freedom (DoF) to grasp objects gives rise to many challenges. In this paper, we propose a novel imitation learning approach to learn a policy that directly predicts 23 DoF grasp trajectories from a partial point cloud provided by a single, fixed camera. At the core of the approach is a second-order geometric-based model of behavioral dynamics. This Neural Geometric Fabric (NGF) policy predicts accelerations directly in joint space. We show that our policy is capable of generalizing to novel objects, and combine our policy with a geometric fabric motion planner in a loop to generate stable grasping trajectories. We evaluate our approach on a set of three different objects, compare different policy structures, and run ablation studies to understand the importance of different object encodings for policy learning.

Robotic assembly for high-mixture settings requires adaptivity to diverse parts and poses, which is an open challenge. Meanwhile, in other areas of robotics, large models and sim-to-real have led to tremendous progress. Inspired by such work, we present AutoMate, a learning framework and system that consists of 4 parts: 1) a dataset of 100 assemblies compatible with simulation and the real world, along with parallelized simulation environments for policy learning, 2) a novel simulation-based approach for learning specialist (i.e., part-specific) policies and generalist (i.e., unified) assembly policies, 3) demonstrations of specialist policies that individually solve 80 assemblies with 80 a generalist policy that jointly solves 20 assemblies with an 80 rate, and 4) zero-shot sim-to-real transfer that achieves similar (or better) performance than simulation, including on perception-initialized assembly. The key methodological takeaway is that a union of diverse algorithms from manufacturing engineering, character animation, and time-series analysis provides a generic and robust solution for a diverse range of robotic assembly problems.To our knowledge, AutoMate provides the first simulation-based framework for learning specialist and generalist policies over a wide range of assemblies, as well as the first system demonstrating zero-shot sim-to-real transfer over such a range.

Most dynamics functions are not well-aligned to task requirements. Controllers, therefore, often invert the dynamics and reshape it into something more useful. The learning community has found that these controllers, such as Operational Space Control (OSC), can offer important inductive biases for training. However, OSC only captures straight line end-effector motion. There's a lot more behavior we could and should be packing into these systems. Earlier work [15][16][19] developed a theory that generalized these ideas and constructed a broad and flexible class of second-order dynamical systems which was simultaneously expressive enough to capture substantial behavior (such as that listed above), and maintained the types of stability properties that make OSC and controllers like it a good foundation for policy design and learning. This paper, motivated by the empirical success of the types of fabrics used in [20], reformulates the theory of fabrics into a form that's more general and easier to apply to policy learning problems. We focus on the stability properties that make fabrics a good foundation for policy synthesis. Fabrics create a fundamentally stable medium within which a policy can operate; they influence the system's behavior without preventing it from achieving tasks within its constraints. When a fabrics is geometric (path consistent) we can interpret the fabric as forming a road network of paths that the system wants to follow at constant speed absent a forcing policy, giving geometric intuition to its role as a prior. The policy operating over the geometric fabric acts to modulate speed and steers the system from one road to the next as it accomplishes its task. We reformulate the theory of fabrics here rigorously and develop theoretical results characterizing system behavior and illuminating how to design these systems, while also emphasizing intuition throughout.

Motion generation seeks to produce safe and feasible robot motion from start to goal. Various tools at different levels of granularity have been developed. On one extreme, sampling-based motion planners focus on completeness - a solution, if it exists, would eventually be found. However, produced paths are often of low quality, and contain superfluous motion. On the other, reactive methods optimise the immediate cost to obtain the next controls, producing smooth and legible motion that can quickly adapt to perturbations, uncertainties, and changes in the environment. However, reactive methods are highly local, and often produce motion that become trapped in non-convex regions of the environment. This paper contributes, Geometric Fabric Command Sequences, a method that lies in the middle ground. It can produce globally optimal motion that is smooth and intuitive, while being also reactive. We model motion via a reactive Geometric Fabric policy that ingests a sequence of attractor states, or commands, and then apply global optimisation over the space of commands. We postulate that solutions for different problems and scenes are highly transferable when conditioned on environmental features. Therefore, an implicit generative model is trained on solutions from optimisation and environment features in a self-supervised manner. That is, faced with multiple motion generation problems, the learning and optimisation are contained within the same loop: the optimisation generates labels for learning, while the learning improves the optimisation for the next problem, which in turn provides higher quality labels. We empirically validate our method in both simulation and on a real-world 6-DOF JACO arm.

Vision-based teleoperation offers the possibility to endow robots with human-level intelligence to physically interact with the environment, while only requiring low-cost camera sensors. However, current vision-based teleoperation systems are designed and engineered towards a particular robot model and deploy environment, which scales poorly as the pool of the robot models expands and the variety of the operating environment increases. In this paper, we propose AnyTeleop, a unified and general teleoperation system to support multiple different arms, hands, realities, and camera configurations within a single system. Although being designed to provide great flexibility to the choice of simulators and real hardware, our system can still achieve great performance. For real-world experiments, AnyTeleop can outperform a previous system that was designed for a specific robot hardware with a higher success rate, using the same robot. For teleoperation in simulation, AnyTeleop leads to better imitation learning performance, compared with a previous system that is particularly designed for that simulator. Project page: http://anyteleop.com/.

This paper explores the problem of collision-free motion generation for manipulators by formulating it as a global motion optimization problem. We develop a parallel optimization technique to solve this problem and demonstrate its effectiveness on massively parallel GPUs. We show that combining simple optimization techniques with many parallel seeds leads to solving difficult motion generation problems within 50ms on average, 60x faster than state-of-the-art (SOTA) trajectory optimization methods. We achieve SOTA performance by combining L-BFGS step direction estimation with a novel parallel noisy line search scheme and a particle-based optimization solver. To further aid trajectory optimization, we develop a parallel geometric planner that plans within 20ms and also introduce a collision-free IK solver that can solve over 7000 queries/s. We package our contributions into a state of the art GPU accelerated motion generation library, cuRobo and release it to enrich the robotics community. Additional details are available at https://curobo.org

This paper explores the problem of collision-free motion generation for manipulators by formulating it as a global motion optimization problem. We develop a parallel optimization technique to solve this problem and demonstrate its effectiveness on massively parallel GPUs. We show that combining simple optimization techniques with many parallel seeds leads to solving difficult motion generation problems within 53ms on average, 62x faster than SOTA trajectory optimization methods. We achieve SOTA performance by combining L-BFGS step direction estimation with a novel parallel noisy line search scheme and a particle-based optimization solver. To further aid trajectory optimization, we develop a parallel geometric planner that is atleast 28x faster than SOTA RRTConnect implementations. We also introduce a collision-free IK solver that can solve over 9000 queries/s. We are releasing our GPU accelerated library CuRobo that contains core components for robot motion generation. Additional details are available at sites.google.com/nvidia.com/curobo.

Recent work has demonstrated the ability of deep reinforcement learning (RL) algorithms to learn complex robotic behaviours in simulation, including in the domain of multi-fingered manipulation. However, such models can be challenging to transfer to the real world due to the gap between simulation and reality. In this paper, we present our techniques to train a) a policy that can perform robust dexterous manipulation on an anthropomorphic robot hand and b) a robust pose estimator suitable for providing reliable real-time information on the state of the object being manipulated. Our policies are trained to adapt to a wide range of conditions in simulation. Consequently, our vision-based policies significantly outperform the best vision policies in the literature on the same reorientation task and are competitive with policies that are given privileged state information via motion capture systems. Our work reaffirms the possibilities of sim-to-real transfer for dexterous manipulation in diverse kinds of hardware and simulator setups, and in our case, with the Allegro Hand and Isaac Gym GPU-based simulation. Furthermore, it opens up possibilities for researchers to achieve such results with commonly-available, affordable robot hands and cameras. Videos of the resulting policy and supplementary information, including experiments and demos, can be found on the website.

Dexterous robotic hands have the capability to interact with a wide variety of household objects to perform tasks like grasping. However, learning robust real world grasping policies for arbitrary objects has proven challenging due to the difficulty of generating high quality training data. In this work, we propose a learning system (ISAGrasp) for leveraging a small number of human demonstrations to bootstrap the generation of a much larger dataset containing successful grasps on a variety of novel objects. Our key insight is to use a correspondence-aware implicit generative model to deform object meshes and demonstrated human grasps in order to generate a diverse dataset of novel objects and successful grasps for supervised learning, while maintaining semantic realism. We use this dataset to train a robust grasping policy in simulation which can be deployed in the real world. We demonstrate grasping performance with a four-fingered Allegro hand in both simulation and the real world, and show this method can handle entirely new semantic classes and achieve a 79% success rate on grasping unseen objects in the real world.

Classical mechanical systems are central to controller design in energy shaping methods of geometric control. However, their expressivity is limited by position-only metrics and the intimate link between metric and geometry. Recent work on Riemannian Motion Policies (RMPs) has shown that shedding these restrictions results in powerful design tools, but at the expense of theoretical stability guarantees. In this work, we generalize classical mechanics to what we call geometric fabrics, whose expressivity and theory enable the design of systems that outperform RMPs in practice. Geometric fabrics strictly generalize classical mechanics forming a new physics of behavior by first generalizing them to Finsler geometries and then explicitly bending them to shape their behavior while maintaining stability. We develop the theory of fabrics and present both a collection of controlled experiments examining their theoretical properties and a set of robot system experiments showing improved performance over a well-engineered and hardened implementation of RMPs, our current state-of-the-art in controller design.

Learning dexterous manipulation policies for multi-fingered robots has been a long-standing challenge in robotics. Existing methods either limit themselves to highly constrained problems and smaller models to achieve extreme sample efficiency or sacrifice sample efficiency to gain capacity to solve more complex tasks with deep neural networks. In this work, we develop a structured approach to sample-efficient learning of dexterous manipulation skills from demonstrations by leveraging recent advances in robot motion generation and control. Specifically, our policy structure is induced by Geometric Fabrics - a recent framework that generalizes classical mechanical systems to allow for flexible design of expressive robot motions. To avoid the cumbersome manual design required by existing motion generators, we introduce Neural Geometric Fabric (NGF) - a framework that learns Geometric Fabric-based policies from data. NGF policies are provably stable and capable of encoding speed-invariant geometries of complex motions in multiple task spaces simultaneously. We demonstrate that NGFs can learn to perform a variety of dexterous manipulation tasks on a 23-DoF hand-arm physical robotic platform purely from demonstrations. Results from comprehensive comparative and ablative experiments show that NGF's structure and action spaces help learn acceleration-based policies that consistently outperform state-of-the-art baselines like Riemannian Motion Policies (RMPs), and other commonly used networks, such as feed-forward and recurrent neural networks. More importantly, we demonstrate that NGFs do not rely on often-used and expertly-designed operational-space controllers, promoting an advancement towards efficiently learning safe, stable, and high-dimensional controllers.