CoRL 2023 论文列表

We propose a novel end-to-end pipeline for online long-range vectorized high-definition (HD) map construction using on-board camera sensors. The vectorized representation of HD maps, employing polylines and polygons to represent map elements, is widely used by downstream tasks. However, previous schemes designed with reference to dynamic object detection overlook the structural constraints within linear map elements, resulting in performance degradation in long-range scenarios. In this paper, we exploit the properties of map elements to improve the performance of map construction. We extract more accurate bird’s eye view (BEV) features guided by their linear structure, and then propose a hierarchical sparse map representation to further leverage the scalability of vectorized map elements, and design a progressive decoding mechanism and a supervision strategy based on this representation. Our approach, ScalableMap, demonstrates superior performance on the nuScenes dataset, especially in long-range scenarios, surpassing previous state-of-the-art model by 6.5 mAP while achieving 18.3 FPS.

This paper presents a data-driven strategy to streamline the deployment of model-based controllers in legged robotic hardware platforms. Our approach leverages a model-free safe learning algorithm to automate the tuning of control gains, addressing the mismatch between the simplified model used in the control formulation and the real system. This method substantially mitigates the risk of hazardous interactions with the robot by sample-efficiently optimizing parameters within a probably safe region. Additionally, we extend the applicability of our approach to incorporate the different gait parameters as contexts, leading to a safe, sample-efficient exploration algorithm capable of tuning a motion controller for diverse gait patterns. We validate our method through simulation and hardware experiments, where we demonstrate that the algorithm obtains superior performance on tuning a model-based motion controller for multiple gaits safely.

Multi-agent reinforcement learning (MARL) has emerged as a popular technique in diverse domains due to its ability to automate system controller design and facilitate continuous intelligence learning. For instance, traffic flow is often trained with MARL to enable intelligent simulations for autonomous driving. However, The existing MARL algorithm only characterizes the relative degree of each agent's contribution to the team, and cannot express the contribution that the team needs from the agent. Especially in the field of autonomous driving, the team changes over time, and the agent needs to act directly according to the needs of the team. To address these limitations, we propose an innovative method inspired by realistic traffic coordinators called the Traffic Coordinator Network (TraCo). Our approach leverages a combination of cross-attention and counterfactual advantage function, allowing us to extract distinctive characteristics of domain agents and accurately quantify the contribution that a team needs from an agent. Through experiments conducted on four traffic tasks, we demonstrate that our method outperforms existing approaches, yielding superior performance. Furthermore, our approach enables the emergence of rich and diverse social behaviors among vehicles within the traffic flow.

We focus on developing efficient and reliable policy optimization strategies for robot learning with real-world data. In recent years, policy gradient methods have emerged as a promising paradigm for training control policies in simulation. However, these approaches often remain too data inefficient or unreliable to train on real robotic hardware. In this paper we introduce a novel policy gradient-based policy optimization framework which systematically leverages a (possibly highly simplified) first-principles model and enables learning precise control policies with limited amounts of real-world data. Our approach $1)$ uses the derivatives of the model to produce sample-efficient estimates of the policy gradient and $2)$ uses the model to design a low-level tracking controller, which is embedded in the policy class. Theoretical analysis provides insight into how the presence of this feedback controller addresses overcomes key limitations of stand-alone policy gradient methods, while hardware experiments with a small car and quadruped demonstrate that our approach can learn precise control strategies reliably and with only minutes of real-world data.

Large language models (LLMs) have demonstrated exciting progress in acquiring diverse new capabilities through in-context learning, ranging from logical reasoning to code-writing. Robotics researchers have also explored using LLMs to advance the capabilities of robotic control. However, since low-level robot actions are hardware-dependent and underrepresented in LLM training corpora, existing efforts in applying LLMs to robotics have largely treated LLMs as semantic planners or relied on human-engineered control primitives to interface with the robot. On the other hand, reward functions are shown to be flexible representations that can be optimized for control policies to achieve diverse tasks, while their semantic richness makes them suitable to be specified by LLMs. In this work, we introduce a new paradigm that harnesses this realization by utilizing LLMs to define reward parameters that can be optimized and accomplish variety of robotic tasks. Using reward as the intermediate interface generated by LLMs, we can effectively bridge the gap between high-level language instructions or corrections to low-level robot actions. Meanwhile, combining this with a real-time optimizer, MuJoCo MPC, empowers an interactive behavior creation experience where users can immediately observe the results and provide feedback to the system. To systematically evaluate the performance of our proposed method, we designed a total of 17 tasks for a simulated quadruped robot and a dexterous manipulator robot. We demonstrate that our proposed method reliably tackles 90% of the designed tasks, while a baseline using primitive skills as the interface with Code-as-policies achieves 50% of the tasks. We further validated our method on a real robot arm where complex manipulation skills such as non-prehensile pushing emerge through our interactive system.

We observe that pre-trained large language models (LLMs) are capable of autoregressively completing complex token sequences—from arbitrary ones procedurally generated by probabilistic context-free grammars (PCFG), to more rich spatial patterns found in the Abstraction and Reasoning Corpus (ARC), a general AI benchmark, prompted in the style of ASCII art. Surprisingly, pattern completion proficiency can be partially retained even when the sequences are expressed using tokens randomly sampled from the vocabulary. These results suggest that without any additional training, LLMs can serve as general sequence modelers, driven by in-context learning. In this work, we investigate how these zero-shot capabilities may be applied to problems in robotics—from extrapolating sequences of numbers that represent states over time to complete simple motions, to least-to-most prompting of reward-conditioned trajectories that can discover and represent closed-loop policies (e.g., a stabilizing controller for CartPole). While difficult to deploy today for real systems due to latency, context size limitations, and compute costs, the approach of using LLMs to drive low-level control may provide an exciting glimpse into how the patterns among words could be transferred to actions.

Self-supervised and language-supervised image models contain rich knowledge of the world that is important for generalization. Many robotic tasks, however, require a detailed understanding of 3D geometry, which is often lacking in 2D image features. This work bridges this 2D-to-3D gap for robotic manipulation by leveraging distilled feature fields to combine accurate 3D geometry with rich semantics from 2D foundation models. We present a few-shot learning method for 6-DOF grasping and placing that harnesses these strong spatial and semantic priors to achieve in-the-wild generalization to unseen objects. Using features distilled from a vision-language model, CLIP, we present a way to designate novel objects for manipulation via free-text natural language, and demonstrate its ability to generalize to unseen expressions and novel categories of objects. Project website: https://f3rm.csail.mit.edu

Learning robot policies from few demonstrations is crucial in open-ended applications. We propose a new method, Interaction Warping, for one-shot learning SE(3) robotic manipulation policies. We infer the 3D mesh of each object in the environment using shape warping, a technique for aligning point clouds across object instances. Then, we represent manipulation actions as keypoints on objects, which can be warped with the shape of the object. We show successful one-shot imitation learning on three simulated and real-world object re-arrangement tasks. We also demonstrate the ability of our method to predict object meshes and robot grasps in the wild. Webpage: https://shapewarping.github.io.

To plan efficient robot locomotion, we must use the information about a terrain’s physics that can be inferred from color images. To this end, we train a visual perception module that predicts terrain properties using labels from a small amount of real-world proprioceptive locomotion. To ensure label precision, we introduce Active Sensing Motor Policies (ASMP). These policies are trained to prefer motor skills that facilitate accurately estimating the environment’s physics, like swiping a foot to observe friction. The estimated labels supervise a vision model that infers physical properties directly from color images and can be reused for different tasks. Leveraging a pretrained vision backbone, we demonstrate robust generalization in image space, enabling path planning from overhead imagery despite using only ground camera images for training.

We introduce Rotateit, a system that enables fingertip-based object rotation along multiple axes by leveraging multimodal sensory inputs. Our system is trained in simulation, where it has access to ground-truth object shapes and physical properties. Then we distill it to operate on realistic yet noisy simulated visuotactile and proprioceptive sensory inputs. These multimodal inputs are fused via a visuotactile transformer, enabling online inference of object shapes and physical properties during deployment. We show significant performance improvements over prior methods and highlight the importance of visual and tactile sensing.

Imitation learning is a powerful tool for training robot manipulation policies, allowing them to learn from expert demonstrations without manual programming or trial-and-error. However, common methods of data collection, such as human supervision, scale poorly, as they are time-consuming and labor-intensive. In contrast, Task and Motion Planning (TAMP) can autonomously generate large-scale datasets of diverse demonstrations. In this work, we show that the combination of large-scale datasets generated by TAMP supervisors and flexible Transformer models to fit them is a powerful paradigm for robot manipulation. We present a novel imitation learning system called OPTIMUS that trains large-scale visuomotor Transformer policies by imitating a TAMP agent. We conduct a thorough study of the design decisions required to imitate TAMP and demonstrate that OPTIMUS can solve a wide variety of challenging vision-based manipulation tasks with over 70 different objects, ranging from long-horizon pick-and-place tasks, to shelf and articulated object manipulation, achieving 70 to 80% success rates. Video results and code at https://mihdalal.github.io/optimus/

Learning complex locomotion and manipulation tasks presents significant challenges, often requiring extensive engineering of, e.g., reward functions or curricula to provide meaningful feedback to the Reinforcement Learning (RL) algorithm. This paper proposes an intrinsically motivated RL approach to reduce task-specific engineering. The desired task is encoded in a single sparse reward, i.e., a reward of “+1” is given if the task is achieved. Intrinsic motivation enables learning by guiding exploration toward the sparse reward signal. Specifically, we adapt the idea of Random Network Distillation (RND) to the robotics domain to learn holistic motion control policies involving simultaneous locomotion and manipulation. We investigate opening doors as an exemplary task for robotic ap- plications. A second task involving package manipulation from a table to a bin highlights the generalization capabilities of the presented approach. Finally, the resulting RL policies are executed in real-world experiments on a wheeled-legged robot in biped mode. We experienced no failure in our experiments, which consisted of opening push doors (over 15 times in a row) and manipulating packages (over 5 times in a row).

To deploy autonomous vehicles(AVs) in the real world, developers must understand the conditions in which the system can operate safely. To do this in a scalable manner, AVs are often tested in simulation on parameterized scenarios. In this context, it’s important to build a testing framework that partitions the scenario parameter space into safe, unsafe, and unknown regions. Existing approaches rely on discretizing continuous parameter spaces into bins, which scales poorly to high-dimensional spaces and cannot describe regions with arbitrary shape. In this work, we introduce a problem formulation which avoids discretization — by modeling the probability of meeting safety requirements everywhere, the parameter space can be paritioned using a probability threshold. Based on our formulation, we propose GUARD as a testing framework which leverages Gaussian Processes to model probability and levelset algorithms to efficiently generate tests. Moreover, we introduce a set of novel evaluation metrics for coverage-based testing frameworks to capture the key objectives of testing. In our evaluation suite of diverse high-dimensional scenarios, GUARD significantly outperforms existing approaches. By proposing an efficient, accurate, and scalable testing framework, our work is a step towards safely deploying autonomous vehicles at scale.

A rich representation is key to general robotic manipulation, but existing approaches to representation learning require large amounts of multimodal demonstrations. In this work we propose PLEX, a transformer-based architecture that learns from a small amount of task-agnostic visuomotor trajectories and a much larger amount of task-conditioned object manipulation videos — a type of data available in quantity. PLEX uses visuomotor trajectories to induce a latent feature space and to learn task-agnostic manipulation routines, while diverse video-only demonstrations teach PLEX how to plan in the induced latent feature space for a wide variety of tasks. Experiments showcase PLEX’s generalization on Meta-World and SOTA performance in challenging Robosuite environments. In particular, using relative positional encoding in PLEX’s transformers greatly helps in low-data regimes of learning from human-collected demonstrations.

Imitation learning is a paradigm to address complex motion planning problems by learning a policy to imitate an expert's behavior. However, relying solely on the expert's data might lead to unsafe actions when the robot deviates from the demonstrated trajectories. Stability guarantees have previously been provided utilizing nonlinear dynamical systems, acting as high-level motion planners, in conjunction with the Lyapunov stability theorem. Yet, these methods are prone to inaccurate policies, high computational cost, sample inefficiency, or quasi stability when replicating complex and highly nonlinear trajectories. To mitigate this problem, we present an approach for learning a globally stable nonlinear dynamical system as a motion planning policy. We model the nonlinear dynamical system as a parametric polynomial and learn the polynomial's coefficients jointly with a Lyapunov candidate. To showcase its success, we compare our method against the state of the art in simulation and conduct real-world experiments with the Kinova Gen3 Lite manipulator arm. Our experiments demonstrate the sample efficiency and reproduction accuracy of our method for various expert trajectories, while remaining stable in the face of perturbations.

Humans use different modalities, such as speech, text, images, videos, etc., to communicate their intent and goals with teammates. For robots to become better assistants, we aim to endow them with the ability to follow instructions and understand tasks specified by their human partners. Most robotic policy learning methods have focused on one single modality of task specification while ignoring the rich cross-modal information. We present MUTEX, a unified approach to policy learning from multimodal task specifications. It trains a transformer-based architecture to facilitate cross-modal reasoning, combining masked modeling and cross-modal matching objectives in a two-stage training procedure. After training, MUTEX can follow a task specification in any of the six learned modalities (video demonstrations, goal images, text goal descriptions, text instructions, speech goal descriptions, and speech instructions) or a combination of them. We systematically evaluate the benefits of MUTEX in a newly designed dataset with 100 tasks in simulation and 50 tasks in the real world, annotated with multiple instances of task specifications in different modalities, and observe improved performance over methods trained specifically for any single modality. More information at https://ut-austin-rpl.github.io/MUTEX/

Navigation in unfamiliar environments presents a major challenge for robots: while mapping and planning techniques can be used to build up a representation of the world, quickly discovering a path to a desired goal in unfamiliar settings with such methods often requires lengthy mapping and exploration. Humans can rapidly navigate new environments, particularly indoor environments that are laid out logically, by leveraging semantics --- e.g., a kitchen often adjoins a living room, an exit sign indicates the way out, and so forth. Language models can provide robots with such knowledge, but directly using language models to instruct a robot how to reach some destination can also be impractical: while language models might produce a narrative about how to reach some goal, because they are not grounded in real-world observations, this narrative might be arbitrarily wrong. Therefore, in this paper we study how the ``semantic guesswork'' produced by language models can be utilized as a guiding heuristic for planning algorithms. Our method, Language Frontier Guide (LFG), uses the language model to bias exploration of novel real-world environments by incorporating the semantic knowledge stored in language models as a search heuristic for planning with either topological or metric maps. We evaluate LFG in challenging real-world environments and simulated benchmarks, outperforming uninformed exploration and other ways of using language models.

This paper introduces a novel approach for modeling continuous forward kinematic models of soft continuum robots by employing Augmented Neural ODE (ANODE), a cutting-edge family of deep neural network models. To the best of our knowledge, this is the first application of ANODE in modeling soft continuum robots. This formulation introduces auxiliary dimensions, allowing the system's states to evolve in the augmented space which provides a richer set of dynamics that the model can learn, increasing the flexibility and accuracy of the model. Our methodology achieves exceptional sample efficiency, training the continuous forward kinematic model using only 25 scattered data points. Additionally, we design and implement a fully parallel Model Predictive Path Integral (MPPI)-based controller running on a GPU, which efficiently manages a non-convex objective function. Through a set of experiments, we showed that the proposed framework (ANODE+MPPI) significantly outperforms state-of-the-art learning-based methods such as FNN and RNN in unseen-before scenarios and marginally outperforms them in seen-before scenarios.

Simulation forms the backbone of modern self-driving development. Simulators help develop, test, and improve driving systems without putting humans, vehicles, or their environment at risk. However, simulators face a major challenge: They rely on realistic, scalable, yet interesting content. While recent advances in rendering and scene reconstruction make great strides in creating static scene assets, modeling their layout, dynamics, and behaviors remains challenging. In this work, we turn to language as a source of supervision for dynamic traffic scene generation. Our model, LCTGen, combines a large language model with a transformer-based decoder architecture that selects likely map locations from a dataset of maps, and produces an initial traffic distribution, as well as the dynamics of each vehicle. LCTGen outperforms prior work in both unconditional and conditional traffic scene generation in terms of realism and fidelity.