Robot learning is witnessing a significant increase in the size, diversity, and complexity of pre-collected datasets, mirroring trends in domains such as natural language processing and computer vision. Many robot learning methods treat such datasets as multi-task expert data and learn a multi-task, generalist policy by training broadly across them. Notably, while these generalist policies can improve the average performance across many tasks, the performance of generalist policies on any one task is often suboptimal due to negative transfer between partitions of the data, compared to task-specific specialist policies. In this work, we argue for the paradigm of training policies during deployment given the scenarios they encounter: rather than deploying pre-trained policies to unseen problems in a zero-shot manner, we non-parametrically retrieve and train models directly on relevant data at test time. Furthermore, we show that many robotics tasks share considerable amounts of low-level behaviors and that retrieval at the "sub"-trajectory granularity enables significantly improved data utilization, generalization, and robustness in adapting policies to novel problems. In contrast, existing full-trajectory retrieval methods tend to underutilize the data and miss out on shared cross-task content. This work proposes STRAP, a technique for leveraging pre-trained vision foundation models and dynamic time warping to retrieve sub-sequences of trajectories from large training corpora in a robust fashion. STRAP outperforms both prior retrieval algorithms and multi-task learning methods in simulated and real experiments, showing the ability to scale to much larger offline datasets in the real world as well as the ability to learn robust control policies with just a handful of real-world demonstrations.

Simulation is an indispensable tool in the development and testing of autonomous vehicles (AVs), offering an efficient and safe alternative to road testing. An outstanding challenge with simulation-based testing is the generation of safety-critical scenarios, which are essential to ensure that AVs can handle rare but potentially fatal situations. This paper addresses this challenge by introducing a novel framework CaDRE, to generate realistic, diverse, and controllable safetycritical scenarios. Our approach optimizes for both the quality and diversity of scenarios by employing a unique formulation and algorithm that integrates real-world scenarios, domain knowledge, and black-box optimization. We validate the effectiveness of our framework through extensive testing in three representative types of traffic scenarios. The results demonstrate superior performance in generating diverse and highquality scenarios with greater sample efficiency than existing reinforcement learning (RL) and sampling-based methods.

Various pose estimation and tracking problems in robotics can be decomposed into a correspondence estimation problem (often computed using a deep network) fol-lowed by a weighted least squares optimization problem to solve for the poses. Recent work has shown that coupling the two problems by iteratively refining one conditioned on the other's output yields SOTA results across domains. However, training these models has proved challenging, re-quiring a litany of tricks to stabilize and speed up training. In this work, we take the visual odometry problem as an example and identify three plausible causes: (1) flow loss interference, (2) linearization errors in the bundle adjust-ment (BA) layer, and (3) dependence of weight gradients on the BA residual. We show how these issues result in noisy and higher variance gradients, potentially leading to a slow down in training and instabilities. We then propose a sim-ple, yet effective solution to reduce the gradient variance by using the weights predicted by the network in the inner opti-mization loop to weight the correspondence objective in the training problem. This helps the training objective 'focus' on the more important points, thereby reducing the variance and mitigating the influence of outliers. We show that the resulting method leads to faster training and can be more flexibly trained in varying training setups without sacrificing performance. In particular we show 2-2.5x training speedups over a baseline visual odometry model we modify.

Recognizing the potential of freehand gestures for interacting with virtual objects in virtual environments, our research introduces a user-defined freehand gesture set of typical referents in virtual reality (VR), focusing on a specific scenario: VR museums. We conducted a comprehensive elicitation study with two experiments to define and refine the gesture set. Meanwhile, we demonstrated an enhanced real-time Wizard of Oz approach that facilitated users’ understanding of referents in VR and their gesture design. Our findings revealed significant improvements in gesture consistency and user agreement through two experiments, with an average agreement score of firstchoice and second-choice advancing from 0.211 and 0.160 to 0.412 and 0.284, respectively. By offering a consistent user-centered gesture set, this work contributes to guiding museum curators toward creating more immersive user experiences in VR museums. The gestures can also be extended to other VR applications that necessitate travel, selection and manipulation, and system control tasks.

Anomaly detection has many important applications, such as monitoring industrial equipment. Despite recent advances in anomaly detection with deep-learning methods, it is unclear how existing solutions would perform under out-of-distribution scenarios, e.g., due to shifts in machine load or environmental noise. Grounded in the application of machine health monitoring, we propose a framework that adapts to new conditions with few-shot samples. Building upon prior work, we adopt a classification-based approach for anomaly detection and show its equivalence to mixture density estimation of the normal samples. We incorporate an episodic training procedure to match the few-shot setting during inference. We define multiple auxiliary classification tasks based on meta-information and leverage gradient-based meta-learning to improve generalization to different shifts. We evaluate our proposed method on a recently-released dataset of audio measurements from different machine types. It improved upon two baselines by around 10% and is on par with best-performing model reported on the dataset.

Existing research on autonomous driving primarily focuses on urban driving, which is insufficient for characterising the complex driving behaviour underlying high-speed racing. At the same time, existing racing simulation frameworks struggle in capturing realism, with respect to visual rendering, vehicular dynamics, and task objectives, inhibiting the transfer of learning agents to real-world contexts. We introduce a new environment, where agents Learn-to-Race (L2R) in simulated competition-style racing, using multimodal information—from virtual cameras to a comprehensive array of inertial measurement sensors. Our environment, which includes a simulator and an interfacing training framework, accurately models vehicle dynamics and racing conditions. In this paper, we release the Arrival simulator for autonomous racing. Next, we propose the L2R task with challenging metrics, inspired by learning-to-drive challenges, Formula-style racing, and multimodal trajectory prediction for autonomous driving. Additionally, we provide the L2R framework suite, facilitating simulated racing on high-precision models of real-world tracks. Finally, we provide an official L2R task dataset of expert demonstrations, as well as a series of baseline experiments and reference implementations. We make all code available: https://github.com/learn-to-race/l2r.

Non-intrusive load monitoring (NILM) is the set of algorithmic techniques for inferring the operational states of individual appliances in a household given the aggregate electrical measurements at a single point of instrumentation. Most successful techniques to-date approach the problem from a supervised learning perspective and thus rely on labeled data, which is costly to obtain, and assume similar data distributions for appliances beyond those in the training set. To alleviate this problem, we formulated NILM in a domain adaptation context. Using Binary OnLine FactorizaTion (BOLT) as the baseline model, we first demonstrate that direct application of domain adversarial training without application-specific modifications is unsuccessful, and hypothesize that this may be due to differences in the data distributions. We then propose Dyna-BOLT: a domain-adaptive variant of BOLT, in which we 1) provide private decoders for the source and target domains to account for the differences in current waveforms, and 2) tie the weights between the two decoders using a metric that was specifically trained to distinguish between appliance classes. We evaluate Dyna-BOLT on a publicly available dataset (REDD) and demonstrate that it compares favorably to unsupervised methods.

We introduce a physics-guided signal processing approach to extract a damage-sensitive and domain-invariant (DS & DI) feature from acceleration response data of a vehicle traveling over a bridge to assess bridge health. Motivated by indirect sensing methods' benefits, such as low-cost and low-maintenance, vehicle-vibration-based bridge health monitoring has been studied to efficiently monitor bridges in real-time. Yet applying this approach is challenging because 1) physics-based features extracted manually are generally not damage-sensitive, and 2) features from machine learning techniques are often not applicable to different bridges. Thus, we formulate a vehicle bridge interaction system model and find a physics-guided DS & DI feature, which can be extracted using the synchrosqueezed wavelet transform representing non-stationary signals as intrinsic-mode-type components. We validate the effectiveness of the proposed feature with simulated experiments. Compared to conventional time-and frequency-domain features, our feature provides the best damage quantification and localization results across different bridges in five of six experiments.