David Hoeller

Mayank Mittal, Calvin Yu, Qinxi Yu et al. · cmu, eth-zurich

We present Orbit, a unified and modular framework for robot learning powered by NVIDIA Isaac Sim. It offers a modular design to easily and efficiently create robotic environments with photo-realistic scenes and high-fidelity rigid and deformable body simulation. With Orbit, we provide a suite of benchmark tasks of varying difficulty -- from single-stage cabinet opening and cloth folding to multi-stage tasks such as room reorganization. To support working with diverse observations and action spaces, we include fixed-arm and mobile manipulators with different physically-based sensors and motion generators. Orbit allows training reinforcement learning policies and collecting large demonstration datasets from hand-crafted or expert solutions in a matter of minutes by leveraging GPU-based parallelization. In summary, we offer an open-sourced framework that readily comes with 16 robotic platforms, 4 sensor modalities, 10 motion generators, more than 20 benchmark tasks, and wrappers to 4 learning libraries. With this framework, we aim to support various research areas, including representation learning, reinforcement learning, imitation learning, and task and motion planning. We hope it helps establish interdisciplinary collaborations in these communities, and its modularity makes it easily extensible for more tasks and applications in the future.

Mayank Mittal, Pascal Roth, James Tigue et al. · nvidia

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.

Nikita Rudin, David Hoeller, Marko Bjelonic et al. · eth-zurich

The common approach for local navigation on challenging environments with legged robots requires path planning, path following and locomotion, which usually requires a locomotion control policy that accurately tracks a commanded velocity. However, by breaking down the navigation problem into these sub-tasks, we limit the robot's capabilities since the individual tasks do not consider the full solution space. In this work, we propose to solve the complete problem by training an end-to-end policy with deep reinforcement learning. Instead of continuously tracking a precomputed path, the robot needs to reach a target position within a provided time. The task's success is only evaluated at the end of an episode, meaning that the policy does not need to reach the target as fast as possible. It is free to select its path and the locomotion gait. Training a policy in this way opens up a larger set of possible solutions, which allows the robot to learn more complex behaviors. We compare our approach to velocity tracking and additionally show that the time dependence of the task reward is critical to successfully learn these new behaviors. Finally, we demonstrate the successful deployment of policies on a real quadrupedal robot. The robot is able to cross challenging terrains, which were not possible previously, while using a more energy-efficient gait and achieving a higher success rate.

David Hoeller, Nikita Rudin, Christopher Choy et al. · eth-zurich

We propose a learning-based method to reconstruct the local terrain for locomotion with a mobile robot traversing urban environments. Using a stream of depth measurements from the onboard cameras and the robot's trajectory, the algorithm estimates the topography in the robot's vicinity. The raw measurements from these cameras are noisy and only provide partial and occluded observations that in many cases do not show the terrain the robot stands on. Therefore, we propose a 3D reconstruction model that faithfully reconstructs the scene, despite the noisy measurements and large amounts of missing data coming from the blind spots of the camera arrangement. The model consists of a 4D fully convolutional network on point clouds that learns the geometric priors to complete the scene from the context and an auto-regressive feedback to leverage spatio-temporal consistency and use evidence from the past. The network can be solely trained with synthetic data, and due to extensive augmentation, it is robust in the real world, as shown in the validation on a quadrupedal robot, ANYmal, traversing challenging settings. We run the pipeline on the robot's onboard low-power computer using an efficient sparse tensor implementation and show that the proposed method outperforms classical map representations.

Clemens Schwarke, Mayank Mittal, Nikita Rudin et al.

RSL-RL is an open-source Reinforcement Learning library tailored to the specific needs of the robotics community. Unlike broad general-purpose frameworks, its design philosophy prioritizes a compact and easily modifiable codebase, allowing researchers to adapt and extend algorithms with minimal overhead. The library focuses on algorithms most widely adopted in robotics, together with auxiliary techniques that address robotics-specific challenges. Optimized for GPU-only training, RSL-RL achieves high-throughput performance in large-scale simulation environments. Its effectiveness has been validated in both simulation benchmarks and in real-world robotic experiments, demonstrating its utility as a lightweight, extensible, and practical framework to develop learning-based robotic controllers. The library is open-sourced at: https://github.com/leggedrobotics/rsl_rl.

Nikita Rudin, David Hoeller, Philipp Reist et al.

In this work, we present and study a training set-up that achieves fast policy generation for real-world robotic tasks by using massive parallelism on a single workstation GPU. We analyze and discuss the impact of different training algorithm components in the massively parallel regime on the final policy performance and training times. In addition, we present a novel game-inspired curriculum that is well suited for training with thousands of simulated robots in parallel. We evaluate the approach by training the quadrupedal robot ANYmal to walk on challenging terrain. The parallel approach allows training policies for flat terrain in under four minutes, and in twenty minutes for uneven terrain. This represents a speedup of multiple orders of magnitude compared to previous work. Finally, we transfer the policies to the real robot to validate the approach. We open-source our training code to help accelerate further research in the field of learned legged locomotion.

Viktor Makoviychuk, Lukasz Wawrzyniak, Yunrong Guo et al.

Isaac Gym offers a high performance learning platform to train policies for wide variety of robotics tasks directly on GPU. Both physics simulation and the neural network policy training reside on GPU and communicate by directly passing data from physics buffers to PyTorch tensors without ever going through any CPU bottlenecks. This leads to blazing fast training times for complex robotics tasks on a single GPU with 2-3 orders of magnitude improvements compared to conventional RL training that uses a CPU based simulator and GPU for neural networks. We host the results and videos at \url{https://sites.google.com/view/isaacgym-nvidia} and isaac gym can be downloaded at \url{https://developer.nvidia.com/isaac-gym}.

Mayank Mittal, David Hoeller, Farbod Farshidian et al.

A kitchen assistant needs to operate human-scale objects, such as cabinets and ovens, in unmapped environments with dynamic obstacles. Autonomous interactions in such environments require integrating dexterous manipulation and fluid mobility. While mobile manipulators in different form factors provide an extended workspace, their real-world adoption has been limited. Executing a high-level task for general objects requires a perceptual understanding of the object as well as adaptive whole-body control among dynamic obstacles. In this paper, we propose a two-stage architecture for autonomous interaction with large articulated objects in unknown environments. The first stage, object-centric planner, only focuses on the object to provide an action-conditional sequence of states for manipulation using RGB-D data. The second stage, agent-centric planner, formulates the whole-body motion control as an optimal control problem that ensures safe tracking of the generated plan, even in scenes with moving obstacles. We show that the proposed pipeline can handle complex static and dynamic kitchen settings for both wheel-based and legged mobile manipulators. Compared to other agent-centric planners, our proposed planner achieves a higher success rate and a lower execution time. We also perform hardware tests on a legged mobile manipulator to interact with various articulated objects in a kitchen. For additional material, please check: www.pair.toronto.edu/articulated-mm/.

David Hoeller, Lorenz Wellhausen, Farbod Farshidian et al.

In this work, we present a learning-based pipeline to realise local navigation with a quadrupedal robot in cluttered environments with static and dynamic obstacles. Given high-level navigation commands, the robot is able to safely locomote to a target location based on frames from a depth camera without any explicit mapping of the environment. First, the sequence of images and the current trajectory of the camera are fused to form a model of the world using state representation learning. The output of this lightweight module is then directly fed into a target-reaching and obstacle-avoiding policy trained with reinforcement learning. We show that decoupling the pipeline into these components results in a sample efficient policy learning stage that can be fully trained in simulation in just a dozen minutes. The key part is the state representation, which is trained to not only estimate the hidden state of the world in an unsupervised fashion, but also helps bridging the reality gap, enabling successful sim-to-real transfer. In our experiments with the quadrupedal robot ANYmal in simulation and in reality, we show that our system can handle noisy depth images, avoid dynamic obstacles unseen during training, and is endowed with local spatial awareness.

Visak Kumar, David Hoeller, Balakumar Sundaralingam et al.

The dominant way to control a robot manipulator uses hand-crafted differential equations leveraging some form of inverse kinematics / dynamics. We propose a simple, versatile joint-level controller that dispenses with differential equations entirely. A deep neural network, trained via model-free reinforcement learning, is used to map from task space to joint space. Experiments show the method capable of achieving similar error to traditional methods, while greatly simplifying the process by automatically handling redundancy, joint limits, and acceleration / deceleration profiles. The basic technique is extended to avoid obstacles by augmenting the input to the network with information about the nearest obstacles. Results are shown both in simulation and on a real robot via sim-to-real transfer of the learned policy. We show that it is possible to achieve sub-centimeter accuracy, both in simulation and the real world, with a moderate amount of training.

Xingye Da, Zhaoming Xie, David Hoeller et al.

We present a hierarchical framework that combines model-based control and reinforcement learning (RL) to synthesize robust controllers for a quadruped (the Unitree Laikago). The system consists of a high-level controller that learns to choose from a set of primitives in response to changes in the environment and a low-level controller that utilizes an established control method to robustly execute the primitives. Our framework learns a controller that can adapt to challenging environmental changes on the fly, including novel scenarios not seen during training. The learned controller is up to 85~percent more energy efficient and is more robust compared to baseline methods. We also deploy the controller on a physical robot without any randomization or adaptation scheme.

Napat Karnchanachari, Miguel I. Valls, David Hoeller et al.

Model Predictive Control (MPC) is a powerful control technique that handles constraints, takes the system's dynamics into account, and optimizes for a given cost function. In practice, however, it often requires an expert to craft and tune this cost function and find trade-offs between different state penalties to satisfy simple high level objectives. In this paper, we use Reinforcement Learning and in particular value learning to approximate the value function given only high level objectives, which can be sparse and binary. Building upon previous works, we present improvements that allowed us to successfully deploy the method on a real world unmanned ground vehicle. Our experiments show that our method can learn the cost function from scratch and without human intervention, while reaching a performance level similar to that of an expert-tuned MPC. We perform a quantitative comparison of these methods with standard MPC approaches both in simulation and on the real robot.

Farbod Farshidian, David Hoeller, Marco Hutter

In this paper, we introduce an actor-critic algorithm called Deep Value Model Predictive Control (DMPC), which combines model-based trajectory optimization with value function estimation. The DMPC actor is a Model Predictive Control (MPC) optimizer with an objective function defined in terms of a value function estimated by the critic. We show that our MPC actor is an importance sampler, which minimizes an upper bound of the cross-entropy to the state distribution of the optimal sampling policy. In our experiments with a Ballbot system, we show that our algorithm can work with sparse and binary reward signals to efficiently solve obstacle avoidance and target reaching tasks. Compared to previous work, we show that including the value function in the running cost of the trajectory optimizer speeds up the convergence. We also discuss the necessary strategies to robustify the algorithm in practice.