A rich representation is key to general robotic manipulation, but existing model architectures require a lot of data to learn it. Unfortunately, ideal robotic manipulation training data, which comes in the form of expert visuomotor demonstrations for a variety of annotated tasks, is scarce. In this work we propose PLEX, a transformer-based architecture that learns from task-agnostic visuomotor trajectories accompanied by a much larger amount of task-conditioned object manipulation videos -- a type of robotics-relevant data available in quantity. The key insight behind PLEX is that the trajectories with observations and actions help induce a latent feature space and train a robot to execute task-agnostic manipulation routines, while a diverse set of video-only demonstrations can efficiently teach the robot how to plan in this feature space for a wide variety of tasks. In contrast to most works on robotic manipulation pretraining, PLEX learns a generalizable sensorimotor multi-task policy, not just an observational representation. We also show that using relative positional encoding in PLEX's transformers further increases its data efficiency when learning from human-collected demonstrations. Experiments showcase \appr's generalization on Meta-World-v2 benchmark and establish state-of-the-art performance in challenging Robosuite environments.
Spatial understanding is a fundamental aspect of computer vision and integral for human-level reasoning about images, making it an important component for grounded language understanding. While recent large-scale text-to-image synthesis (T2I) models have shown unprecedented improvements in photorealism, it is unclear whether they have reliable spatial understanding capabilities. We investigate the ability of T2I models to generate correct spatial relationships among objects and present VISOR, an evaluation metric that captures how accurately the spatial relationship described in text is generated in the image. To benchmark existing models, we introduce a large-scale challenge dataset SR2D that contains sentences describing two objects and the spatial relationship between them. We construct and harness an automated evaluation pipeline that employs computer vision to recognize objects and their spatial relationships, and we employ it in a large-scale evaluation of T2I models. Our experiments reveal a surprising finding that, although recent state-of-the-art T2I models exhibit high image quality, they are severely limited in their ability to generate multiple objects or the specified spatial relations such as left/right/above/below. Our analyses demonstrate several biases and artifacts of T2I models such as the difficulty with generating multiple objects, a bias towards generating the first object mentioned, spatially inconsistent outputs for equivalent relationships, and a correlation between object co-occurrence and spatial understanding capabilities. We conduct a human study that shows the alignment between VISOR and human judgment about spatial understanding. We offer the SR2D dataset and the VISOR metric to the community in support of T2I spatial reasoning research.
We propose EM-PASTE: an Expectation Maximization(EM) guided Cut-Paste compositional dataset augmentation approach for weakly-supervised instance segmentation using only image-level supervision. The proposed method consists of three main components. The first component generates high-quality foreground object masks. To this end, an EM-like approach is proposed that iteratively refines an initial set of object mask proposals generated by a generic region proposal method. Next, in the second component, high-quality context-aware background images are generated using a text-to-image compositional synthesis method like DALL-E. Finally, the third component creates a large-scale pseudo-labeled instance segmentation training dataset by compositing the foreground object masks onto the original and generated background images. The proposed approach achieves state-of-the-art weakly-supervised instance segmentation results on both the PASCAL VOC 2012 and MS COCO datasets by using only image-level, weak label information. In particular, it outperforms the best baseline by +7.4 and +2.8 mAP0.50 on PASCAL and COCO, respectively. Further, the method provides a new solution to the long-tail weakly-supervised instance segmentation problem (when many classes may only have few training samples), by selectively augmenting under-represented classes.
Simulating realistic sensors is a challenging part in data generation for autonomous systems, often involving carefully handcrafted sensor design, scene properties, and physics modeling. To alleviate this, we introduce a pipeline for data-driven simulation of a realistic LiDAR sensor. We propose a model that learns a mapping between RGB images and corresponding LiDAR features such as raydrop or per-point intensities directly from real datasets. We show that our model can learn to encode realistic effects such as dropped points on transparent surfaces or high intensity returns on reflective materials. When applied to naively raycasted point clouds provided by off-the-shelf simulator software, our model enhances the data by predicting intensities and removing points based on the scene's appearance to match a real LiDAR sensor. We use our technique to learn models of two distinct LiDAR sensors and use them to improve simulated LiDAR data accordingly. Through a sample task of vehicle segmentation, we show that enhancing simulated point clouds with our technique improves downstream task performance.
Training computer vision models usually requires collecting and labeling vast amounts of imagery under a diverse set of scene configurations and properties. This process is incredibly time-consuming, and it is challenging to ensure that the captured data distribution maps well to the target domain of an application scenario. Recently, synthetic data has emerged as a way to address both of these issues. However, existing approaches either require human experts to manually tune each scene property or use automatic methods that provide little to no control; this requires rendering large amounts of random data variations, which is slow and is often suboptimal for the target domain. We present the first fully differentiable synthetic data pipeline that uses Neural Radiance Fields (NeRFs) in a closed-loop with a target application's loss function. Our approach generates data on-demand, with no human labor, to maximize accuracy for a target task. We illustrate the effectiveness of our method on synthetic and real-world object detection tasks. We also introduce a new "YCB-in-the-Wild" dataset and benchmark that provides a test scenario for object detection with varied poses in real-world environments.
Weakly supervised object detection (WSOD) enables object detectors to be trained using image-level class labels. However, the practical application of current WSOD models is limited, as they operate at small scales and require extensive training and refinement. We propose the Weakly Supervised Detection Transformer, which enables efficient knowledge transfer from a large-scale pretraining dataset to WSOD finetuning on hundreds of novel objects. We leverage pretrained knowledge to improve the multiple instance learning framework used in WSOD, and experiments show our approach outperforms the state-of-the-art on datasets with twice the novel classes than previously shown.
3D scene graphs (3DSGs) are an emerging description; unifying symbolic, topological, and metric scene representations. However, typical 3DSGs contain hundreds of objects and symbols even for small environments; rendering task planning on the full graph impractical. We construct TASKOGRAPHY, the first large-scale robotic task planning benchmark over 3DSGs. While most benchmarking efforts in this area focus on vision-based planning, we systematically study symbolic planning, to decouple planning performance from visual representation learning. We observe that, among existing methods, neither classical nor learning-based planners are capable of real-time planning over full 3DSGs. Enabling real-time planning demands progress on both (a) sparsifying 3DSGs for tractable planning and (b) designing planners that better exploit 3DSG hierarchies. Towards the former goal, we propose SCRUB, a task-conditioned 3DSG sparsification method; enabling classical planners to match and in some cases surpass state-of-the-art learning-based planners. Towards the latter goal, we propose SEEK, a procedure enabling learning-based planners to exploit 3DSG structure, reducing the number of replanning queries required by current best approaches by an order of magnitude. We will open-source all code and baselines to spur further research along the intersections of robot task planning, learning and 3DSGs.
Joint visual and language modeling on large-scale datasets has recently shown a good progress in multi-modal tasks when compared to single modal learning. However, robustness of these approaches against real-world perturbations has not been studied. In this work, we perform the first extensive robustness study of such models against various real-world perturbations focusing on video and language. We focus on text-to-video retrieval and propose two large-scale benchmark datasets, MSRVTT-P and YouCook2-P, which utilize 90 different visual and 35 different textual perturbations. The study reveals some interesting findings: 1) The studied models are more robust when text is perturbed versus when video is perturbed 2) The transformer text encoder is more robust on non-semantic changing text perturbations and visual perturbations compared to word embedding approaches. 3) Using two-branch encoders in isolation is typically more robust than when architectures use cross-attention. We hope this study will serve as a benchmark and guide future research in robust multimodal learning.
We have seen a great progress in video action recognition in recent years. There are several models based on convolutional neural network (CNN) with some recent transformer based approaches which provide state-of-the-art performance on existing benchmark datasets. However, large-scale robustness has not been studied for these models which is a critical aspect for real-world applications. In this work we perform a large-scale robustness analysis of these existing models for video action recognition. We mainly focus on robustness against distribution shifts due to real-world perturbations instead of adversarial perturbations. We propose four different benchmark datasets, HMDB-51P, UCF-101P, Kinetics-400P, and SSv2P and study the robustness of six different state-of-the-art action recognition models against 90 different perturbations. The study reveals some interesting findings, 1) transformer based models are consistently more robust against most of the perturbations when compared with CNN based models, 2) Pretraining helps Transformer based models to be more robust to different perturbations than CNN based models, and 3) All of the studied models are robust to temporal perturbation on the Kinetics dataset, but not on SSv2; this suggests temporal information is much more important for action label prediction on SSv2 datasets than on the Kinetics dataset. We hope that this study will serve as a benchmark for future research in robust video action recognition. More details about the project are available at https://rose-ar.github.io/.
Object cut-and-paste has become a promising approach to efficiently generate large sets of labeled training data. It involves compositing foreground object masks onto background images. The background images, when congruent with the objects, provide helpful context information for training object recognition models. While the approach can easily generate large labeled data, finding congruent context images for downstream tasks has remained an elusive problem. In this work, we propose a new paradigm for automatic context image generation at scale. At the core of our approach lies utilizing an interplay between language description of context and language-driven image generation. Language description of a context is provided by applying an image captioning method on a small set of images representing the context. These language descriptions are then used to generate diverse sets of context images using the language-based DALL-E image generation framework. These are then composited with objects to provide an augmented training set for a classifier. We demonstrate the advantages of our approach over the prior context image generation approaches on four object detection datasets. Furthermore, we also highlight the compositional nature of our data generation approach on out-of-distribution and zero-shot data generation scenarios.