The unprecedented surge in video data production in recent years necessitates efficient tools to extract meaningful frames from videos for downstream tasks. Long-term temporal reasoning is a key desideratum for frame retrieval systems. While state-of-the-art foundation models, like VideoLLaMA and ViCLIP, are proficient in short-term semantic understanding, they surprisingly fail at long-term reasoning across frames. A key reason for this failure is that they intertwine per-frame perception and temporal reasoning into a single deep network. Hence, decoupling but co-designing semantic understanding and temporal reasoning is essential for efficient scene identification. We propose a system that leverages vision-language models for semantic understanding of individual frames but effectively reasons about the long-term evolution of events using state machines and temporal logic (TL) formulae that inherently capture memory. Our TL-based reasoning improves the F1 score of complex event identification by 9-15% compared to benchmarks that use GPT4 for reasoning on state-of-the-art self-driving datasets such as Waymo and NuScenes.
Imagine generating a city's electricity demand pattern based on weather, the presence of an electric vehicle, and location, which could be used for capacity planning during a winter freeze. Such real-world time series are often enriched with paired heterogeneous contextual metadata (weather, location, etc.). Current approaches to time series generation often ignore this paired metadata, and its heterogeneity poses several practical challenges in adapting existing conditional generation approaches from the image, audio, and video domains to the time series domain. To address this gap, we introduce Time Weaver, a novel diffusion-based model that leverages the heterogeneous metadata in the form of categorical, continuous, and even time-variant variables to significantly improve time series generation. Additionally, we show that naive extensions of standard evaluation metrics from the image to the time series domain are insufficient. These metrics do not penalize conditional generation approaches for their poor specificity in reproducing the metadata-specific features in the generated time series. Thus, we innovate a novel evaluation metric that accurately captures the specificity of conditional generation and the realism of the generated time series. We show that Time Weaver outperforms state-of-the-art benchmarks, such as Generative Adversarial Networks (GANs), by up to 27% in downstream classification tasks on real-world energy, medical, air quality, and traffic data sets.
Foundation models have recently expanded into robotics after excelling in computer vision and natural language processing. The models are accessible in two ways: open-source or paid, closed-source options. Users with access to both face a problem when deciding between effective yet costly closed-source models and free but less powerful open-source alternatives. We call it the model selection problem. Existing supervised-learning methods are impractical due to the high cost of collecting extensive training data from closed-source models. Hence, we focus on the online learning setting where algorithms learn while collecting data, eliminating the need for large pre-collected datasets. We thus formulate a user-centric online model selection problem and propose a novel solution that combines an open-source encoder to output context and an online learning algorithm that processes this context. The encoder distills vast data distributions into low-dimensional features, i.e., the context, without additional training. The online learning algorithm aims to maximize a composite reward that includes model performance, execution time, and costs based on the context extracted from the data. It results in an improved trade-off between selecting open-source and closed-source models compared to non-contextual methods, as validated by our theoretical analysis. Experiments across language-based robotic tasks such as Waymo Open Dataset, ALFRED, and Open X-Embodiment demonstrate real-world applications of the solution. The results show that the solution significantly improves the task success rate by up to 14%.
The future of work does not require a choice between human and robot. Aside from explicit human-robot collaboration, robotics can play an increasingly important role in helping train workers as well as the tools they may use, especially in complex tasks that may be difficult to automate or effectively roboticize. This paper introduces a form of smart tool for use by human workers and shows how training the tool for task recognition, one of the key requirements, can be accomplished. Machine learning (ML) with purely human-based data can be extremely laborious and time-consuming. First, we show how data synthetically-generated by a robot can be leveraged in the ML training process. Later, we demonstrate how fine-tuning ML models for individual physical tasks and workers can significantly scale up the benefits of using ML to provide this feedback. Experimental results show the effectiveness and scalability of our approach, as we test data size versus accuracy. Smart hand tools of the type introduced here can provide insights and real-time analytics on efficient and safe tool usage and operation, thereby enhancing human participation and skill in a wide range of work environments. Using robotic platforms to help train smart tools will be essential, particularly given the diverse types of applications for which smart hand tools are envisioned for human use.
We present an autonomous navigation system that operates without assuming HD LiDAR maps of the environment. Our system, ALT-Pilot, relies only on publicly available road network information and a sparse (and noisy) set of crowdsourced language landmarks. With the help of onboard sensors and a language-augmented topometric map, ALT-Pilot autonomously pilots the vehicle to any destination on the road network. We achieve this by leveraging vision-language models pre-trained on web-scale data to identify potential landmarks in a scene, incorporating vision-language features into the recursive Bayesian state estimation stack to generate global (route) plans, and a reactive trajectory planner and controller operating in the vehicle frame. We implement and evaluate ALT-Pilot in simulation and on a real, full-scale autonomous vehicle and report improvements over state-of-the-art topometric navigation systems by a factor of 3x on localization accuracy and 5x on goal reachability
The increasing abundance of video data enables users to search for events of interest, e.g., emergency incidents. Meanwhile, it raises new concerns, such as the need for preserving privacy. Existing approaches to video search require either manual inspection or a deep learning model with massive training. We develop a method that uses recent advances in vision and language models, as well as formal methods, to search for events of interest in video clips automatically and efficiently. The method consists of an algorithm to map text-based event descriptions into linear temporal logic over finite traces (LTL$_f$) and an algorithm to construct an automaton encoding the video information. Then, the method formally verifies the automaton representing the video against the LTL$_f$ specifications and adds the pertinent video clips to the search result if the automaton satisfies the specifications. We provide qualitative and quantitative analysis to demonstrate the video-searching capability of the proposed method. It achieves over 90 percent precision in searching over privacy-sensitive videos and a state-of-the-art autonomous driving dataset.
Efficient compression of correlated data is essential to minimize communication overload in multi-sensor networks. In such networks, each sensor independently compresses the data and transmits them to a central node due to limited communication bandwidth. A decoder at the central node decompresses and passes the data to a pre-trained machine learning-based task to generate the final output. Thus, it is important to compress the features that are relevant to the task. Additionally, the final performance depends heavily on the total available bandwidth. In practice, it is common to encounter varying availability in bandwidth, and higher bandwidth results in better performance of the task. We design a novel distributed compression framework composed of independent encoders and a joint decoder, which we call neural distributed principal component analysis (NDPCA). NDPCA flexibly compresses data from multiple sources to any available bandwidth with a single model, reducing computing and storage overhead. NDPCA achieves this by learning low-rank task representations and efficiently distributing bandwidth among sensors, thus providing a graceful trade-off between performance and bandwidth. Experiments show that NDPCA improves the success rate of multi-view robotic arm manipulation by 9% and the accuracy of object detection tasks on satellite imagery by 14% compared to an autoencoder with uniform bandwidth allocation.
Fleets of networked autonomous vehicles (AVs) collect terabytes of sensory data, which is often transmitted to central servers (the ''cloud'') for training machine learning (ML) models. Ideally, these fleets should upload all their data, especially from rare operating contexts, in order to train robust ML models. However, this is infeasible due to prohibitive network bandwidth and data labeling costs. Instead, we propose a cooperative data sampling strategy where geo-distributed AVs collaborate to collect a diverse ML training dataset in the cloud. Since the AVs have a shared objective but minimal information about each other's local data distribution and perception model, we can naturally cast cooperative data collection as an $N$-player mathematical game. We show that our cooperative sampling strategy uses minimal information to converge to a centralized oracle policy with complete information about all AVs. Moreover, we theoretically characterize the performance benefits of our game-theoretic strategy compared to greedy sampling. Finally, we experimentally demonstrate that our method outperforms standard benchmarks by up to $21.9\%$ on 4 perception datasets, including for autonomous driving in adverse weather conditions. Crucially, our experimental results on real-world datasets closely align with our theoretical guarantees.
Recent progress in deep reinforcement learning (RL) and computer vision enables artificial agents to solve complex tasks, including locomotion, manipulation and video games from high-dimensional pixel observations. However, domain specific reward functions are often engineered to provide sufficient learning signals, requiring expert knowledge. While it is possible to train vision-based RL agents using only sparse rewards, additional challenges in exploration arise. We present a novel and efficient method to solve sparse-reward robot manipulation tasks from only image observations by utilizing a few demonstrations. First, we learn an embedded neural dynamics model from demonstration transitions and further fine-tune it with the replay buffer. Next, we reward the agents for staying close to the demonstrated trajectories using a distance metric defined in the embedding space. Finally, we use an off-policy, model-free vision RL algorithm to update the control policies. Our method achieves state-of-the-art sample efficiency in simulation and enables efficient training of a real Franka Emika Panda manipulator.
Robotic perception models, such as Deep Neural Networks (DNNs), are becoming more computationally intensive and there are several models being trained with accuracy and latency trade-offs. However, modern latency accuracy trade-offs largely report mean accuracy for single-step vision tasks, but there is little work showing which model to invoke for multi-step control tasks in robotics. The key challenge in a multi-step decision making is to make use of the right models at right times to accomplish the given task. That is, the accomplishment of the task with a minimum control cost and minimum perception time is a desideratum; this is known as the model selection problem. In this work, we precisely address this problem of invoking the correct sequence of perception models for multi-step control. In other words, we provide a provably optimal solution to the model selection problem by casting it as a multi-objective optimization problem balancing the control cost and perception time. The key insight obtained from our solution is how the variance of the perception models matters (not just the mean accuracy) for multi-step decision making, and to show how to use diverse perception models as a primitive for energy-efficient robotics. Further, we demonstrate our approach on a photo-realistic drone landing simulation using visual navigation in AirSim. Using our proposed policy, we achieved 38.04% lower control cost with 79.1% less perception time than other competing benchmarks.