CoRL 2024 Papers

Robot decision-making increasingly relies on data-driven human prediction models when operating around people. While these models are known to mispredict in out-of-distribution interactions, only a subset of prediction errors impact downstream robot performance. We propose characterizing such ``system-level'' prediction failures via the mathematical notion of regret: high-regret interactions are precisely those in which mispredictions degraded closed-loop robot performance. We further introduce a probabilistic generalization of regret that calibrates failure detection across disparate deployment contexts and renders regret compatible with reward-based and reward-free (e.g., generative) planners. In simulated autonomous driving interactions, we showcase that our system-level failure metric can automatically mine for closed-loop human-robot interactions that state-of-the-art generative human predictors and robot planners struggle with. We further find that the very presence of high-regret data during human predictor fine-tuning is highly predictive of robot re-deployment performance improvements. Furthermore, fine-tuning with the informative but significantly smaller high-regret data (23% of deployment data) is competitive with fine-tuning on the full deployment dataset, indicating a promising avenue for efficiently mitigating system-level human-robot interaction failures.

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.

Imitation learning policies in robotics tend to require an extensive amount of demonstrations. It is critical to develop few-shot adaptation strategies that rely only on a small amount of task-specific human demonstrations. Prior works focus on learning general policies from large scale dataset with diverse behaviors. Recent research has shown that directly retrieving relevant past experiences to augment policy learning has great promise in few-shot settings. However, existing data retrieval methods fall under two extremes: they either rely on the existence of exact same behaviors with visually similar scenes in the prior data, which is impractical to assume; or they retrieve based on semantic similarity of high-level language descriptions of the task, which might not be that informative about the shared behaviors or motions across tasks. In this work, we investigate how we can leverage motion similarity in the vast amount of cross-task data to improve few-shot imitation learning of the target task. Our key insight is that motion-similar data carry rich information about the effects of actions and object interactions that can be leveraged during few-shot adaptation. We propose FlowRetrieval, an approach that leverages optical flow representations for both extracting similar motions to target tasks from prior data, and for guiding learning of a policy that can maximally benefit from such data. Our results show FlowRetrieval significantly outperforms prior methods across simulated and real-world domains, achieving on average 27% higher success rate than the best retrieval-based prior method. In the Pen-in-Cup task with a real Franka Emika robot, FlowRetrieval achieves 3.7x the performance of the baseline learning from all prior and target data.

Recent work has demonstrated that a promising strategy for teaching robots a wide range of complex skills is by training them on a curriculum of progressively more challenging environments. However, developing an effective curriculum of environment distributions currently requires significant expertise, which must be repeated for every new domain. Our key insight is that environments are often naturally represented as code. Thus, we probe whether effective environment curriculum design can be achieved and automated via code generation by large language models (LLM). In this paper, we introduce Eurekaverse, an unsupervised environment design algorithm that uses LLMs to sample progressively more challenging, diverse, and learnable environments for skill training. We validate Eurekaverse's effectiveness in the domain of quadrupedal parkour learning, in which a quadruped robot must traverse through a variety of obstacle courses. The automatic curriculum designed by Eurekaverse enables gradual learning of complex parkour skills in simulation and can successfully transfer to the real-world, outperforming manual training courses designed by humans.

By combining differentiable rendering with explicit point-based scene representations, 3D Gaussian Splatting (3DGS) has demonstrated breakthrough 3D reconstruction capabilities. However, to date 3DGS has had limited impact on robotics, where high-speed egomotion is pervasive: Egomotion introduces motion blur and leads to artifacts in existing frame-based 3DGS reconstruction methods. To address this challenge, we introduce Event3DGS, an event-based 3DGS framework. By exploiting the exceptional temporal resolution of event cameras, Event3GDS can reconstruct high-fidelity 3D structure and appearance under high-speed egomotion. Extensive experiments on multiple synthetic and real-world datasets demonstrate the superiority of Event3DGS compared with existing event-based dense 3D scene reconstruction frameworks; Event3DGS substantially improves reconstruction quality (+3dB) while reducing computational costs by 95\%. Our framework also allows one to incorporate a few motion-blurred frame-based measurements into the reconstruction process to further improve appearance fidelity without loss of structural accuracy.

For everyday service robotics, the ability to navigate back and forth based on tasks in multi-subscene environments and perform delicate manipulations is crucial and highly practical. While existing robotics primarily focus on complex tasks within a single scene or simple tasks across scalable scenes individually, robots consisting of a mobile base with a robotic arm face the challenge of efficiently representing multiple subscenes, coordinating the collaboration between the mobile base and the robotic arm, and managing delicate tasks in scalable environments. To address this issue, we propose Target-driven Multi-subscene Mobile Manipulation (\textit{TaMMa}), which efficiently handles mobile base movement and fine-grained manipulation across subscenes. Specifically, we obtain a reliable 3D Gaussian initialization of the whole scene using a sparse 3D point cloud with encoded semantics. Through querying the coarse Gaussians, we acquire the approximate pose of the target, navigate the mobile base to approach it, and reduce the scope of precise target pose estimation to the corresponding subscene. Optimizing while moving, we employ diffusion-based depth completion to optimize fine-grained Gaussians and estimate the target's refined pose. For target-driven manipulation, we adopt Gaussians inpainting to obtain precise poses for the origin and destination of the operation in a \textit{think before you do it} manner, enabling fine-grained manipulation. We conduct various experiments on a real robotic to demonstrate our method in effectively and efficiently achieving precise operation tasks across multiple tabletop subscenes.

Lidar-based perception pipelines rely on 3D object detection models to interpret complex scenes. While multiple representations for lidar exist, the range view is enticing since it losslessly encodes the entire lidar sensor output. In this work, we achieve state-of-the-art amongst range view 3D object detection models without using multiple techniques proposed in past range view literature. We explore range view 3D object detection across two modern datasets with substantially different properties: Argoverse 2 and Waymo Open. Our investigation reveals key insights: (1) input feature dimensionality significantly influences the overall performance, (2) surprisingly, employing a classification loss grounded in 3D spatial proximity works as well or better compared to more elaborate IoU-based losses, and (3) addressing non-uniform lidar density via a straightforward range subsampling technique outperforms existing multi-resolution, range-conditioned networks. Our experiments reveal that techniques proposed in recent range view literature are not needed to achieve state-of-the-art performance. Combining the above findings, we establish a new state-of-the-art model for range view 3D object detection — improving AP by 2.2% on the Waymo Open dataset while maintaining a runtime of 10 Hz. We are the first to benchmark a range view model on the Argoverse 2 dataset and outperform strong voxel-based baselines. All models are multi-class and open-source. Code is available at https://github.com/benjaminrwilson/range-view-3d-detection.

A central challenge towards developing robots that can relate human language to their perception and actions is the scarcity of natural language annotations in diverse robot datasets. Moreover, robot policies that follow natural language instructions are typically trained on either templated language or expensive human-labeled instructions, hindering their scalability. To this end, we introduce NILS: Natural language Instruction Labeling for Scalability. NILS automatically labels uncurated, long-horizon robot data at scale in a zero-shot manner without any human intervention. NILS combines pre-trained vision-language foundation models in a sophisticated, carefully considered manner in order to detect objects in a scene, detect object-centric changes, segment tasks from large datasets of unlabelled interaction data and ultimately label behavior datasets. Evaluations on BridgeV2 and a kitchen play dataset show that NILS is able to autonomously annotate diverse robot demonstrations of unlabeled and unstructured datasets, while alleviating several shortcomings of crowdsourced human annotations.

In visibility-based pursuit-evasion tasks, a team of mobile pursuer robots with limited sensing capabilities is tasked with detecting all evaders in a multiply-connected planar environment, whose map may or may not be known to pursuers beforehand. This requires tight coordination among multiple agents to ensure that the omniscient and potentially arbitrarily fast evaders are guaranteed to be detected by the pursuers. Whereas existing methods typically rely on a relatively large team of agents to clear the environment, we propose ViPER, a neural solution that leverages a graph attention network to learn a coordinated yet distributed policy via multi-agent reinforcement learning (MARL). We experimentally demonstrate that ViPER significantly outperforms other state-of-the-art non-learning planners, showcasing its emergent coordinated behaviors and adaptability to more challenging scenarios and various team sizes, and finally deploy its learned policies on hardware in an aerial search task.

We introduce BiGym, a new benchmark and learning environment for mobile bi-manual demo-driven robotic manipulation. BiGym features 40 diverse tasks set in home environments, ranging from simple target reaching to complex kitchen cleaning. To capture the real-world performance accurately, we provide human-collected demonstrations for each task, reflecting the diverse modalities found in real-world robot trajectories. BiGym supports a variety of observations, including proprioceptive data and visual inputs such as RGB, and depth from 3 camera views. To validate the usability of BiGym, we thoroughly benchmark the state-of-the-art imitation learning algorithms and demo-driven reinforcement learning algorithms within the environment and discuss the future opportunities.

Animals have the ability to use their arms and legs for both locomotion and manipulation. We envision quadruped robots to have the same versatility. This work presents a system that empowers a quadruped robot to perform object interactions with its legs, drawing inspiration from non-prehensile manipulation techniques. The proposed system has two main components: a visual manipulation policy module and a loco-manipulator module. The visual manipulation policy module decides how the leg should interact with the object, trained with reinforcement learning (RL) with point cloud observations and object-centric actions. The loco-manipulator controller controls the leg movements and body pose adjustments, implemented based on impedance control and Model Predictive Control (MPC). Besides manipulating objects with a single leg, the proposed system can also select from left or right legs based on the critic maps and move the object to distant goals through robot base adjustment. In the experiments, we evaluate the proposed system with the object pose alignment tasks both in simulation and in the real world, demonstrating object manipulation skills with legs more versatile than previous work.

Autonomous robots will need the ability to make task and motion plans that involve long sequences of actions, e.g. to prepare a meal. One challenge is that the feasibility of actions late in the plan may depend on much earlier actions. This issue is exacerbated if these dependencies exist at a purely geometric level, making them difficult to express for a task planner. Backtracking is a common technique to resolve such geometric dependencies, but its time complexity limits its applicability to short-horizon dependencies. We propose an approach to account for these dependencies by learning a search heuristic for task and motion planning. We evaluate our approach on five quasi-static simulated domains and show a substantial improvement in success rate over the baselines.

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.

Parkour poses a significant challenge for legged robots, requiring navigation through complex environments with agility and precision based on limited sensory inputs. In this work, we introduce a novel method for training end-to-end visual policies, from depth pixels to robot control commands, to achieve agile and safe quadruped locomotion. We formulate robot parkour as a constrained reinforcement learning (RL) problem designed to maximize the emergence of agile skills within the robot's physical limits while ensuring safety. We first train a policy without vision using privileged information about the robot's surroundings. We then generate experience from this privileged policy to warm-start a sample efficient off-policy RL algorithm from depth images. This allows the robot to adapt behaviors from this privileged experience to visual locomotion while circumventing the high computational costs of RL directly from pixels. We demonstrate the effectiveness of our method on a real Solo-12 robot, showcasing its capability to perform a variety of parkour skills such as walking, climbing, leaping, and crawling.

Recent advancements in large-scale multi-task robot learning offer the potential for deploying robot fleets in household and industrial settings, enabling them to perform diverse tasks across various environments. However, AI-enabled robots often face challenges with generalization and robustness when exposed to real-world variability and uncertainty. We introduce Sirius-Fleet, a multi-task interactive robot fleet learning framework to address these challenges. Sirius-Fleet monitors robot performance during deployment and involves humans to correct the robot's actions when necessary. We employ a visual world model to predict the outcomes of future actions and build anomaly predictors to predict whether they will likely result in anomalies. As the robot autonomy improves, the anomaly predictors automatically adapt their prediction criteria, leading to fewer requests for human intervention and gradually reducing human workload over time. Evaluations on large-scale benchmarks demonstrate Sirius-Fleet's effectiveness in improving multi-task policy performance and monitoring accuracy. We demonstrate Sirius-Fleet's performance in both RoboCasa in simulation and Mutex in the real world, two diverse, large-scale multi-task benchmarks. More information is available on the project website: https://ut-austin-rpl.github.io/sirius-fleet

Humanoid activities involving sequential contacts are crucial for complex robotic interactions and operations in the real world and are traditionally solved by model-based motion planning, which is time-consuming and often relies on simplified dynamics models. Although model-free reinforcement learning (RL) has become a powerful tool for versatile and robust whole-body humanoid control, it still requires tedious task-specific tuning and state machine design and suffers from long-horizon exploration issues in tasks involving contact sequences. In this work, we propose WoCoCo (Whole-Body Control with Sequential Contacts), a unified framework to learn whole-body humanoid control with sequential contacts by naturally decomposing the tasks into separate contact stages. Such decomposition facilitates simple and general policy learning pipelines through task-agnostic reward and sim-to-real designs, requiring only one or two task-related terms to be specified for each task. We demonstrated that end-to-end RL-based controllers trained with WoCoCo enable four challenging whole-body humanoid tasks involving diverse contact sequences in the real world without any motion priors: 1) versatile parkour jumping, 2) box loco-manipulation, 3) dynamic clap-and-tap dancing, and 4) cliffside climbing. We further show that WoCoCo is a general framework beyond humanoid by applying it in 22-DoF dinosaur robot loco-manipulation tasks. Website: lecar-lab.github.io/wococo/.

In vision-based behaviour cloning (BC), traditional image-level augmentation methods such as pixel shifting enhance in-domain performance but often struggle with visual domain shifts, including distractors, occlusion, and changes in lighting and backgrounds. Conversely, superimposition-based augmentation, proven effective in computer vision, improves model generalisability by blending training images and out-of-domain images. Despite its potential, the applicability of these methods to vision-based BC remains unclear due to the unique challenges posed by BC demonstrations; specifically, preserving task-critical scene semantics, spatial-temporal relationships, and agent-target interactions is crucial. To address this, we introduce RoboSaGA, a context-aware approach that dynamically adjusts augmentation intensity per pixel based on input saliency derived from the policy. This method ensures aggressive augmentation within task-trivial areas without compromising task-critical information. Furthermore, RoboSaGA seamlessly integrates into existing network architectures without the need for structural changes or additional learning objectives. Our empirical evaluations across both simulated and real-world settings demonstrate that RoboSaGA not only maintains in-domain performance but significantly improves resilience to distractors and background variations.

We introduce SonicSense, a holistic design of hardware and software to enable rich robot object perception through in-hand acoustic vibration sensing. While previous studies have shown promising results with acoustic sensing for object perception, current solutions are constrained to a handful of objects with simple geometries and homogeneous materials, single-finger sensing, and mixing training and testing on the same objects. SonicSense enables container inventory status differentiation, heterogeneous material prediction, 3D shape reconstruction, and object re-identification from a diverse set of 83 real-world objects. Our system employs a simple but effective heuristic exploration policy to interact with the objects as well as end-to-end learning-based algorithms to fuse vibration signals to infer object properties. Our framework underscores the significance of in-hand acoustic vibration sensing in advancing robot tactile perception.

Bimanual manipulation is critical to many robotics applications. In contrast to single-arm manipulation, bimanual manipulation tasks are challenging due to higher-dimensional action spaces. Prior works leverage large amounts of data and primitive actions to address this problem, but may suffer from sample inefficiency and limited generalization across various tasks. To this end, we propose VoxAct-B, a language-conditioned, voxel-based method that leverages Vision Language Models (VLMs) to prioritize key regions within the scene and reconstruct a voxel grid. We provide this voxel grid to our bimanual manipulation policy to learn acting and stabilizing actions. This approach enables more efficient policy learning from voxels and is generalizable to different tasks. In simulation, we show that VoxAct-B outperforms strong baselines on fine-grained bimanual manipulation tasks. Furthermore, we demonstrate VoxAct-B on real-world $\texttt{Open Drawer}$ and $\texttt{Open Jar}$ tasks using two UR5s. Code, data, and videos are available at https://voxact-b.github.io.