| Total: 265
Vision-based robot policy learning, which maps visual inputs to actions, necessitates a holistic understanding of diverse visual tasks beyond singletask needs like classification or segmentation. Inspired by this, we introduce Theia, a vision foundation model for robot learning that distills multiple off-the-shelf vision foundation models trained on varied vision tasks. Theia’s rich visual representations encode diverse visual knowledge, enhancing downstream robot learning. Extensive experiments demonstrate that Theia outperforms its teacher models and prior robot learning models using less training data and smaller model sizes. Additionally, we quantify the quality of pre-trained visual representations and hypothesize that higher entropy in feature norm distributions leads to improved robot learning performance.
In recent years, the transformer architecture has become the de-facto standard for machine learning algorithms applied to natural language processing and computer vision. Despite notable evidence of successful deployment of this architecture in the context of robot learning, we claim that vanilla transformers do not fully exploit the structure of the robot learning problem. We propose Body Transformer (BoT), an architecture that exploits the robot embodiment by providing an inductive bias that guides the learning process. We represent the robot body as a graph of sensors and actuators, and rely on masked attention to pool information through the architecture. The resulting architecture outperforms the vanilla transformer, as well as the classical multilayer perceptron, with respect to task completion, scaling properties, and computational efficiency when representing either imitation or reinforcement learning policies.
Despite the impressive recent advances in learning-based robot control, ensuring robustness to out-of-distribution conditions remains an open challenge. Safety filters can, in principle, keep arbitrary control policies from incurring catastrophic failures by overriding unsafe actions, but existing solutions for complex (e.g., legged) robot dynamics do not span the full motion envelope and instead rely on local, reduced-order models. These filters tend to overly restrict agility and can still fail when perturbed away from nominal conditions. This paper presents the gameplay filter, a new class of predictive safety filter that continually plays out hypothetical matches between its simulation-trained safety strategy and a virtual adversary co-trained to invoke worst-case events and sim-to-real error, and precludes actions that would cause failures down the line. We demonstrate the scalability and robustness of the approach with a first-of-its-kind full-order safety filter for (36-D) quadrupedal dynamics. Physical experiments on two different quadruped platforms demonstrate the superior zero-shot effectiveness of the gameplay filter under large perturbations such as tugging and unmodeled terrain. Experiment videos and open-source software are available online: https://saferobotics.org/research/gameplay-filter
Imitation learning from human demonstrations has shown impressive performance in robotics. However, most results focus on table-top manipulation, lacking the mobility and dexterity necessary for generally useful tasks. In this work, we develop a system for imitating mobile manipulation tasks that are bimanual and require whole-body control. We first present Mobile ALOHA, a low-cost and whole-body teleoperation system for data collection. It augments the ALOHA system with a mobile base, and a whole-body teleoperation interface. Using data collected with Mobile ALOHA, we then perform supervised behavior cloning and find that co-training with existing static ALOHA datasets boosts performance on mobile manipulation tasks. With 50 demonstrations for each task, co-training can increase success rates by up to 90\%, allowing Mobile ALOHA to autonomously complete complex mobile manipulation tasks such as sauteing and serving a piece of shrimp, opening a two-door wall cabinet to store heavy cooking pots, calling and entering an elevator, and lightly rinsing a used pan using a kitchen faucet. We will open-source all the hardware and software implementations upon publication.
Endowing robot hands with human-level dexterity is a long-lasting research objective. Bi-manual robot piano playing constitutes a task that combines challenges from dynamic tasks, such as generating fast while precise motions, with slower but contact-rich manipulation problems. Although reinforcement learning based approaches have shown promising results in single-task performance, these methods struggle in a multi-song setting. Our work aims to close this gap and, thereby, enable imitation learning approaches for robot piano playing at scale. To this end, we introduce the Robot Piano 1 Million (RP1M) dataset, containing bi-manual robot piano playing motion data of more than one million trajectories. We formulate finger placements as an optimal transport problem, thus, enabling automatic annotation of vast amounts of unlabeled songs. Benchmarking existing imitation learning approaches shows that such approaches reach state-of-the-art robot piano playing performance by leveraging RP1M.
Traditionally, model-based reinforcement learning (MBRL) methods exploit neural networks as flexible function approximators to represent $\textit{a priori}$ unknown environment dynamics. However, training data are typically scarce in practice, and these black-box models often fail to generalize. Modeling architectures that leverage known physics can substantially reduce the complexity of system-identification, but break down in the face of complex phenomena such as contact. We introduce a novel framework for learning semi-structured dynamics models for contact-rich systems which seamlessly integrates structured first principles modeling techniques with black-box auto-regressive models. Specifically, we develop an ensemble of probabilistic models to estimate external forces, conditioned on historical observations and actions, and integrate these predictions using known Lagrangian dynamics. With this semi-structured approach, we can make accurate long-horizon predictions with substantially less data than prior methods. We leverage this capability and propose Semi-Structured Reinforcement Learning ($\texttt{SSRL}$) a simple model-based learning framework which pushes the sample complexity boundary for real-world learning. We validate our approach on a real-world Unitree Go1 quadruped robot, learning dynamic gaits -- from scratch -- on both hard and soft surfaces with just a few minutes of real-world data. Video and code are available at: https://sites.google.com/utexas.edu/ssrl
The ability to grasp objects in-the-wild from open-ended language instructions constitutes a fundamental challenge in robotics. An open-world grasping system should be able to combine high-level contextual with low-level physical-geometric reasoning in order to be applicable in arbitrary scenarios. Recent works exploit the web-scale knowledge inherent in large language models (LLMs) to plan and reason in robotic context, but rely on external vision and action models to ground such knowledge into the environment and parameterize actuation. This setup suffers from two major bottlenecks: a) the LLM's reasoning capacity is constrained by the quality of visual grounding, and b) LLMs do not contain low-level spatial understanding of the world, which is essential for grasping in contact-rich scenarios. In this work we demonstrate that modern vision-language models (VLMs) are capable of tackling such limitations, as they are implicitly grounded and can jointly reason about semantics and geometry. We propose \texttt{OWG}, an open-world grasping pipeline that combines VLMs with segmentation and grasp synthesis models to unlock grounded world understanding in three stages: open-ended referring segmentation, grounded grasp planning and grasp ranking via contact reasoning, all of which can be applied zero-shot via suitable visual prompting mechanisms. We conduct extensive evaluation in cluttered indoor scene datasets to showcase \texttt{OWG}'s robustness in grounding from open-ended language, as well as open-world robotic grasping experiments in both simulation and hardware that demonstrate superior performance compared to previous supervised and zero-shot LLM-based methods.
Aerial manipulation combines the versatility and speed of flying platforms with the functional capabilities of mobile manipulation, which presents significant challenges due to the need for precise localization and control. Traditionally, researchers have relied on off-board perception systems, which are limited to expensive and impractical specially equipped indoor environments. In this work, we introduce a novel platform for autonomous aerial manipulation that exclusively utilizes onboard perception systems. Our platform can perform aerial manipulation in various indoor and outdoor environments without depending on external perception systems. Our experimental results demonstrate the platform's ability to autonomously grasp various objects in diverse settings. This advancement significantly improves the scalability and practicality of aerial manipulation applications by eliminating the need for costly tracking solutions. To accelerate future research, we open source our modern ROS 2 software stack and custom hardware design, making our contributions accessible to the broader research community.
Learning to move is a primary goal for animals and robots, where ensuring safety is often important when optimizing control policies on the embodied systems. For complex tasks such as the control of human or humanoid control, the high-dimensional parameter space adds complexity to the safe optimization effort. Current safe exploration algorithms exhibit inefficiency and may even become infeasible with large high-dimensional input spaces. Furthermore, existing high-dimensional constrained optimization methods neglect safety in the search process. In this paper, we propose High-dimensional Safe Bayesian Optimization with local optimistic exploration (HdSafeBO), a novel approach designed to handle high-dimensional sampling problems under probabilistic safety constraints. We introduce a local optimistic strategy to efficiently and safely optimize the objective function, providing a probabilistic safety guarantee and a cumulative safety violation bound. Through the use of isometric embedding, HdSafeBO addresses problems ranging from a few hundred to several thousand dimensions while maintaining safety guarantees. To our knowledge, HdSafeBO is the first algorithm capable of optimizing the control of high-dimensional musculoskeletal systems with high safety probability. We also demonstrate the real-world applicability of HdSafeBO through its use in the safe online optimization of neural stimulation-induced human motion control.
We present our method, LeLaN, which uses action-free egocentric data to learn robust language-conditioned object navigation. By leveraging the knowledge of large vision and language models and grounding this knowledge using pre-trained segmentation and depth estimation models, we can label in-the-wild data from a variety of indoor and outdoor environments with diverse instructions that capture a range of objects with varied granularity and noise in their descriptions. Leveraging this method to label over 50 hours of data collected in indoor and outdoor environments, including robot observations, YouTube video tours, and human-collected walking data allows us to train a policy that can outperform state-of-the-art methods on the zero-shot object navigation task in both success rate and precision.
Learning from human feedback has gained traction in fields like robotics and natural language processing in recent years. While prior works mostly rely on human feedback in the form of comparisons, language is a preferable modality that provides more informative insights into user preferences. In this work, we aim to incorporate comparative language feedback to iteratively improve robot trajectories and to learn reward functions that encode human preferences. To achieve this goal, we learn a shared latent space that integrates trajectory data and language feedback, and subsequently leverage the learned latent space to improve trajectories and learn human preferences. To the best of our knowledge, we are the first to incorporate comparative language feedback into reward learning. Our simulation experiments demonstrate the effectiveness of the learned latent space and the success of our learning algorithms. We also conduct human subject studies that show our reward learning algorithm achieves a 23.9% higher subjective score on average and is 11.3% more time-efficient compared to preference-based reward learning, underscoring the superior performance of our method. Our website is at https://liralab.usc.edu/comparative-language-feedback/.
Learned language-conditioned robot policies often struggle to effectively adapt to new real-world tasks even when pre-trained across a diverse set of instructions. We propose a novel approach for few-shot adaptation to unseen tasks that exploits the semantic understanding of task decomposition provided by vision-language models (VLMs). Our method, Policy Adaptation via Language Optimization (PALO), combines a handful of demonstrations of a task with proposed language decompositions sampled from a VLM to quickly enable rapid nonparametric adaptation, avoiding the need for a larger fine-tuning dataset. We evaluate PALO on extensive real-world experiments consisting of challenging unseen, long-horizon robot manipulation tasks. We find that PALO is able of consistently complete long-horizon, multi-tier tasks in the real world, outperforming state of the art pre-trained generalist policies, and methods that have access to the same demonstrations.
In complex real-world tasks such as robotic manipulation and autonomous driving, collecting expert demonstrations is often more straightforward than specifying precise learning objectives and task descriptions. Learning from expert data can be achieved through behavioral cloning or by learning a reward function, i.e., inverse reinforcement learning. The latter allows for training with additional data outside the training distribution, guided by the inferred reward function. We propose a novel approach to construct compact and interpretable reward models from automatically selected state features. These inferred rewards have an explicit form and enable the learning of policies that closely match expert behavior by training standard reinforcement learning algorithms from scratch. We validate our method's performance in various robotic environments with continuous and high-dimensional state spaces.
3D visual grounding is crucial for robots, requiring integration of natural language and 3D scene understanding. Traditional methods depend on supervised learning with 3D point clouds are limited by scarce datasets. Recently zero-shot methods leveraging LLMs have been proposed to address the data issue. While effective, these methods often miss detailed scene context, limiting their ability to handle complex queries. In this work, we present VLM-Grounder, a novel framework using vision-language models (VLMs) for zero-shot 3D visual grounding based solely on 2D images. VLM-Grounder dynamically stitches image sequences, employs a grounding and feedback scheme to find the target object, and uses a multi-view ensemble projection to accurately estimate 3D bounding boxes. Experiments on ScanRefer and Nr3D datasets show VLM-Grounder outperforms previous zero-shot methods, achieving 51.6\% Acc@0.25 on ScanRefer and 48.0\% Acc on Nr3D, without relying on 3D geometry or object priors.
This work introduces Mamba Imitation Learning (MaIL), a novel imitation learning (IL) architecture that offers a computationally efficient alternative to state-of-the-art (SoTA) Transformer policies. Transformer-based policies have achieved remarkable results due to their ability in handling human-recorded data with inherently non-Markovian behavior. However, their high performance comes with the drawback of large models that complicate effective training. While state space models (SSMs) have been known for their efficiency, they were not able to match the performance of Transformers. Mamba significantly improves the performance of SSMs and rivals against Transformers, positioning it as an appealing alternative for IL policies. MaIL leverages Mamba as a backbone and introduces a formalism that allows using Mamba in the encoder-decoder structure. This formalism makes it a versatile architecture that can be used as a standalone policy or as part of a more advanced architecture, such as a diffuser in the diffusion process. Extensive evaluations on the LIBERO IL benchmark and three real robot experiments show that MaIL: i) outperforms Transformers in all LIBERO tasks, ii) achieves good performance even with small datasets, iii) is able to effectively process multi-modal sensory inputs, iv) is more robust to input noise compared to Transformers.
A key limitation of learned robot control policies is their inability to generalize outside their training data. Recent works on vision-language-action models (VLAs) have shown that the use of large, internet pre-trained vision-language models as the backbone of learned robot policies can substantially improve their robustness and generalization ability. Yet, one of the most exciting capabilities of large vision-language models in other domains is their ability to reason iteratively through complex problems. Can that same capability be brought into robotics to allow policies to improve performance by reasoning about a given task before acting? Naive use of "chain-of-thought" (CoT) style prompting is significantly less effective with standard VLAs because of the relatively simple training examples that are available to them. Additionally, purely semantic reasoning about sub-tasks, as is common in regular CoT, is insufficient for robot policies that need to ground their reasoning in sensory observations and the robot state. To this end, we introduce Embodied Chain-of-Thought Reasoning (ECoT) for VLAs, in which we train VLAs to perform multiple steps of reasoning about plans, sub-tasks, motions, and visually grounded features like object bounding boxes and end effector positions, before predicting the robot action. We design a scalable pipeline for generating synthetic training data for ECoT on large robot datasets. We demonstrate, that ECoT increases the absolute success rate of OpenVLA, the current strongest open-source VLA policy, by 28\% across challenging generalization tasks, without any additional robot training data. Additionally, ECoT makes it easier for humans to interpret a policy's failures and correct its behavior using natural language.
Learning visuomotor policies for agile quadrotor flight presents significant difficulties, primarily from inefficient policy exploration caused by high-dimensional visual inputs and the need for precise and low-latency control. To address these challenges, we propose a novel approach that combines the performance of Reinforcement Learning (RL) and the sample efficiency of Imitation Learning (IL) in the task of vision-based autonomous drone racing. While RL provides a framework for learning high-performance controllers through trial and error, it faces challenges with sample efficiency and computational demands due to the high dimensionality of visual inputs. Conversely, IL efficiently learns from visual expert demonstrations, but it remains limited by the expert's performance and state distribution. To overcome these limitations, our policy learning framework integrates the strengths of both approaches. Our framework contains three phases: training a teacher policy using RL with privileged state information, distilling it into a student policy via IL, and adaptive fine-tuning via RL. Testing in both simulated and real-world scenarios shows our approach can not only learn in scenarios where RL from scratch fails but also outperforms existing IL methods in both robustness and performance, successfully navigating a quadrotor through a race course using only visual information.
Intelligent robots capable of improving from autonomously collected experience have the potential to transform robot learning: instead of collecting costly teleoperated demonstration data, large-scale deployment of fleets of robots can quickly collect larger quantities of autonomous data useful for training better robot policies. However, autonomous improvement requires solving two key problems: (i) fully automating a scalable data collection procedure that can collect diverse and semantically meaningful robot data and (ii) learning from non-optimal, autonomous data with no human annotations. To this end, we propose a novel approach that addresses these challenges, allowing instruction following policies to improve from autonomously collected data without human supervision. Our framework leverages vision-language models to collect and evaluate semantically meaningful experiences in new environments, and then utilizes a decomposition of instruction following tasks into (semantic) language-conditioned image generation and (non-semantic) goal reaching, which makes it significantly more practical to improve from this autonomously collected data without any human annotations. We carry out extensive experiments in the real world to demonstrate the effectiveness of our approach, and find that in a suite of unseen environments, the robot policy can be improved significantly with autonomously collected data. We open-source the code for our semantic autonomous improvement pipeline, as well as our autonomous dataset of 25K trajectories collected across five tabletop environments: https://soar-autonomous-improvement.github.io
Learning from expert demonstrations is a popular approach to train robotic manipulation policies from limited data. However, imitation learning algorithms require a number of design choices ranging from the input modality, training objective, and 6-DoF end-effector pose representation. Diffusion-based methods have gained popularity as they allow to predict long horizon trajectories and handle multimodal action distributions. Recently, Conditional Flow Matching (CFM) (or Rectified Flow) has been proposed as a more flexible generalization of diffusion models. In this paper we investigate the application of CFM in the context of robotic policy learning, and specifically study the interplay with the other design choices required to build an imitation learning algorithm. We show that CFM gives the best performance when combined with point cloud input observations. Additionally, we study the feasibility of a CFM formulation on the SO(3) manifold and evaluate its suitability with a simplified example. We perform extensive experiments on RLBench which demonstrate that our proposed PointFlowMatch approach achieves a state-of-the-art average success rate of 67.8% over eight tasks, double the performance of the next best method.
Pre-explored Semantic Map, constructed through prior exploration using visual language models (VLMs), has proven effective as a foundational element for training-free robotic applications. However, existing approaches assume the map's accuracy and do not provide effective mechanisms for revising decisions based on incorrect maps. This work introduces Context-Aware Replanning (CARe),, which estimates map uncertainty through confidence scores and multi-view consistency, enabling the agent to revise erroneous decisions stemming from inaccurate maps without additional labels. We demonstrate the effectiveness of our proposed method using two modern map backbones, VLMaps and OpenMask3D, and show significant improvements in performance on object navigation tasks.
In the field of locomotion task of quadruped robots, Blind Policy and Perceptive Policy each have their own advantages and limitations. The Blind Policy relies on preset sensor information and algorithms, suitable for known and structured environments, but it lacks adaptability in complex or unknown environments. The Perceptive Policy uses visual sensors to obtain detailed environmental information, allowing it to adapt to complex terrains, but its effectiveness is limited under occluded conditions, especially when perception fails. Unlike the Blind Policy, the Perceptive Policy is not as robust under these conditions. To address these challenges, we propose a MBC:Multi-Brain collaborative system that incorporates the concepts of Multi-Agent Reinforcement Learning and introduces collaboration between the Blind Policy and the Perceptive Policy. By applying this multi-policy collaborative model to a quadruped robot, the robot can maintain stable locomotion even when the perceptual system is impaired or observational data is incomplete. Our simulations and real-world experiments demonstrate that this system significantly improves the robot's passability and robustness against perception failures in complex environments, validating the effectiveness of multi-policy collaboration in enhancing robotic motion performance.
Reinforcement learning (RL) has shown remarkable proficiency in developing robust control policies for contact-rich applications. However, it typically requires meticulous Markov Decision Process (MDP) designing tailored to each task and robotic platform. This work addresses this challenge by creating a systematic approach to behavior synthesis and control for multi-contact loco-manipulation. We define a task-independent MDP formulation to learn robust RL policies using a single demonstration (per task) generated from a fast model-based trajectory optimization method. Our framework is validated on diverse real-world tasks, such as navigating spring-loaded doors and manipulating heavy dishwashers. The learned behaviors can handle dynamic uncertainties and external disturbances, showcasing recovery maneuvers, such as re-grasping objects during execution. Finally, we successfully transfer the policies to a real robot, demonstrating the approach's practical viability.
Modern machine learning systems rely on large datasets to attain broad generalization, and this often poses a challenge in robotic learning, where each robotic platform and task might have only a small dataset. By training a single policy across many different kinds of robots, a robotic learning method can leverage much broader and more diverse datasets, which in turn can lead to better generalization and robustness. However, training a single policy on multi-robot data is challenging because robots can have widely varying sensors, actuators, and control frequencies. We propose CrossFormer, a scalable and flexible transformer-based policy that can consume data from any embodiment. We train CrossFormer on the largest and most diverse dataset to date, 900K trajectories across 20 different robot embodiments. We demonstrate that the same network weights can control vastly different robots, including single and dual arm manipulation systems, wheeled robots, quadcopters, and quadrupeds. Unlike prior work, our model does not require manual alignment of the observation or action spaces. Extensive experiments in the real world show that our method matches the performance of specialist policies tailored for each embodiment, while also significantly outperforming the prior state of the art in cross-embodiment learning.
Large policies pretrained on a combination of Internet-scale vision-language data and diverse robot demonstrations have the potential to change how we teach robots new skills: rather than training new behaviors from scratch, we can fine-tune such vision-language-action (VLA) models to obtain robust, generalizable policies for visuomotor control. Yet, widespread adoption of VLAs for robotics has been challenging as 1) existing VLAs are largely closed and inaccessible to the public, and 2) prior work fails to explore methods for efficiently fine-tuning VLAs for new tasks, a key component for adoption. Addressing these challenges, we introduce OpenVLA, a 7B-parameter open-source VLA trained on a diverse collection of 970k real-world robot demonstrations. OpenVLA builds on a Llama 2 language model combined with a visual encoder that fuses pretrained features from DINOv2 and SigLIP. As a product of the added data diversity and new model components, OpenVLA demonstrates strong results for generalist manipulation, outperforming closed models such as RT-2-X (55B) by 16.5\% in absolute task success rate across 29 tasks and multiple robot embodiments, with 7x fewer parameters. We further show that we can effectively fine-tune OpenVLA for new settings, with especially strong generalization results in multi-task environments involving multiple objects and strong language grounding abilities, where we outperform expressive from-scratch imitation learning methods such as Diffusion Policy by 20.4\%. We also explore compute efficiency; as a separate contribution, we show that OpenVLA can be fine-tuned on consumer GPUs via modern low-rank adaptation methods and served efficiently via quantization without a hit to downstream success rate. Finally, we release model checkpoints, fine-tuning notebooks, and our PyTorch codebase with built-in support for training VLAs at scale on Open X-Embodiment datasets.
Large, general-purpose robotic policies trained on diverse demonstration datasets have been shown to be remarkably effective both for controlling a variety of robots in a range of different scenes, and for acquiring broad repertoires of manipulation skills. However, the data that such policies are trained on is generally of mixed quality -- not only are human-collected demonstrations unlikely to perform the task perfectly, but the larger the dataset is, the harder it is to curate only the highest quality examples. It also remains unclear how optimal data from one embodiment is for training on another embodiment. In this paper, we present a general and broadly applicable approach that enhances the performance of such generalist robot policies at deployment time by re-ranking their actions according to a value function learned via offline RL. This approach, which we call Value-Guided Policy Steering (V-GPS), is compatible with a wide range of different generalist policies, without needing to fine-tune or even access the weights of the policy. We show that the same value function can improve the performance of five different state-of-the-art policies with different architectures, even though they were trained on distinct datasets, attaining consistent performance improvement on multiple robotic platforms across a total of 12 tasks. Code and videos can be found at: https://nakamotoo.github.io/V-GPS