Abstract:In this paper, we introduce SPA, a novel representation learning framework that emphasizes the importance of 3D spatial awareness in embodied AI. Our approach leverages differentiable neural rendering on multi-view images to endow a vanilla Vision Transformer (ViT) with intrinsic spatial understanding. We present the most comprehensive evaluation of embodied representation learning to date, covering 268 tasks across 8 simulators with diverse policies in both single-task and language-conditioned multi-task scenarios. The results are compelling: SPA consistently outperforms more than 10 state-of-the-art representation methods, including those specifically designed for embodied AI, vision-centric tasks, and multi-modal applications, while using less training data. Furthermore, we conduct a series of real-world experiments to confirm its effectiveness in practical scenarios. These results highlight the critical role of 3D spatial awareness for embodied representation learning. Our strongest model takes more than 6000 GPU hours to train and we are committed to open-sourcing all code and model weights to foster future research in embodied representation learning. Project Page: https://haoyizhu.github.io/spa/.
Abstract:In this study, we explore the influence of different observation spaces on robot learning, focusing on three predominant modalities: RGB, RGB-D, and point cloud. Through extensive experimentation on over 17 varied contact-rich manipulation tasks, conducted across two benchmarks and simulators, we have observed a notable trend: point cloud-based methods, even those with the simplest designs, frequently surpass their RGB and RGB-D counterparts in performance. This remains consistent in both scenarios: training from scratch and utilizing pretraining. Furthermore, our findings indicate that point cloud observations lead to improved policy zero-shot generalization in relation to various geometry and visual clues, including camera viewpoints, lighting conditions, noise levels and background appearance. The outcomes suggest that 3D point cloud is a valuable observation modality for intricate robotic tasks. We will open-source all our codes and checkpoints, hoping that our insights can help design more generalizable and robust robotic models.
Abstract:In contrast to numerous NLP and 2D computer vision foundational models, the learning of a robust and highly generalized 3D foundational model poses considerably greater challenges. This is primarily due to the inherent data variability and the diversity of downstream tasks. In this paper, we introduce a comprehensive 3D pre-training framework designed to facilitate the acquisition of efficient 3D representations, thereby establishing a pathway to 3D foundational models. Motivated by the fact that informative 3D features should be able to encode rich geometry and appearance cues that can be utilized to render realistic images, we propose a novel universal paradigm to learn point cloud representations by differentiable neural rendering, serving as a bridge between 3D and 2D worlds. We train a point cloud encoder within a devised volumetric neural renderer by comparing the rendered images with the real images. Notably, our approach demonstrates the seamless integration of the learned 3D encoder into diverse downstream tasks. These tasks encompass not only high-level challenges such as 3D detection and segmentation but also low-level objectives like 3D reconstruction and image synthesis, spanning both indoor and outdoor scenarios. Besides, we also illustrate the capability of pre-training a 2D backbone using the proposed universal methodology, surpassing conventional pre-training methods by a large margin. For the first time, PonderV2 achieves state-of-the-art performance on 11 indoor and outdoor benchmarks. The consistent improvements in various settings imply the effectiveness of the proposed method. Code and models will be made available at https://github.com/OpenGVLab/PonderV2.
Abstract:In the context of autonomous driving, the significance of effective feature learning is widely acknowledged. While conventional 3D self-supervised pre-training methods have shown widespread success, most methods follow the ideas originally designed for 2D images. In this paper, we present UniPAD, a novel self-supervised learning paradigm applying 3D volumetric differentiable rendering. UniPAD implicitly encodes 3D space, facilitating the reconstruction of continuous 3D shape structures and the intricate appearance characteristics of their 2D projections. The flexibility of our method enables seamless integration into both 2D and 3D frameworks, enabling a more holistic comprehension of the scenes. We manifest the feasibility and effectiveness of UniPAD by conducting extensive experiments on various downstream 3D tasks. Our method significantly improves lidar-, camera-, and lidar-camera-based baseline by 9.1, 7.7, and 6.9 NDS, respectively. Notably, our pre-training pipeline achieves 73.2 NDS for 3D object detection and 79.4 mIoU for 3D semantic segmentation on the nuScenes validation set, achieving state-of-the-art results in comparison with previous methods. The code will be available at https://github.com/Nightmare-n/UniPAD.
Abstract:A key challenge in robotic manipulation in open domains is how to acquire diverse and generalizable skills for robots. Recent research in one-shot imitation learning has shown promise in transferring trained policies to new tasks based on demonstrations. This feature is attractive for enabling robots to acquire new skills and improving task and motion planning. However, due to limitations in the training dataset, the current focus of the community has mainly been on simple cases, such as push or pick-place tasks, relying solely on visual guidance. In reality, there are many complex skills, some of which may even require both visual and tactile perception to solve. This paper aims to unlock the potential for an agent to generalize to hundreds of real-world skills with multi-modal perception. To achieve this, we have collected a dataset comprising over 110,000 \emph{contact-rich} robot manipulation sequences across diverse skills, contexts, robots, and camera viewpoints, all collected \emph{in the real world}. Each sequence in the dataset includes visual, force, audio, and action information, along with a corresponding human demonstration video. We have invested significant efforts in calibrating all the sensors and ensuring a high-quality dataset. The dataset is made publicly available at rh20t.github.io
Abstract:Accurate whole-body multi-person pose estimation and tracking is an important yet challenging topic in computer vision. To capture the subtle actions of humans for complex behavior analysis, whole-body pose estimation including the face, body, hand and foot is essential over conventional body-only pose estimation. In this paper, we present AlphaPose, a system that can perform accurate whole-body pose estimation and tracking jointly while running in realtime. To this end, we propose several new techniques: Symmetric Integral Keypoint Regression (SIKR) for fast and fine localization, Parametric Pose Non-Maximum-Suppression (P-NMS) for eliminating redundant human detections and Pose Aware Identity Embedding for jointly pose estimation and tracking. During training, we resort to Part-Guided Proposal Generator (PGPG) and multi-domain knowledge distillation to further improve the accuracy. Our method is able to localize whole-body keypoints accurately and tracks humans simultaneously given inaccurate bounding boxes and redundant detections. We show a significant improvement over current state-of-the-art methods in both speed and accuracy on COCO-wholebody, COCO, PoseTrack, and our proposed Halpe-FullBody pose estimation dataset. Our model, source codes and dataset are made publicly available at https://github.com/MVIG-SJTU/AlphaPose.
Abstract:Neural Radiance Fields (NeRFs), despite their outstanding performance on novel view synthesis, often need dense input views. Many papers train one model for each scene respectively and few of them explore incorporating multi-modal data into this problem. In this paper, we focus on a rarely discussed but important setting: can we train one model that can represent multiple scenes, with 360$^\circ $ insufficient views and RGB-D images? We refer insufficient views to few extremely sparse and almost non-overlapping views. To deal with it, X-NeRF, a fully explicit approach which learns a general scene completion process instead of a coordinate-based mapping, is proposed. Given a few insufficient RGB-D input views, X-NeRF first transforms them to a sparse point cloud tensor and then applies a 3D sparse generative Convolutional Neural Network (CNN) to complete it to an explicit radiance field whose volumetric rendering can be conducted fast without running networks during inference. To avoid overfitting, besides common rendering loss, we apply perceptual loss as well as view augmentation through random rotation on point clouds. The proposed methodology significantly out-performs previous implicit methods in our setting, indicating the great potential of proposed problem and approach. Codes and data are available at https://github.com/HaoyiZhu/XNeRF.
Abstract:Autonomous agents have made great strides in specialist domains like Atari games and Go. However, they typically learn tabula rasa in isolated environments with limited and manually conceived objectives, thus failing to generalize across a wide spectrum of tasks and capabilities. Inspired by how humans continually learn and adapt in the open world, we advocate a trinity of ingredients for building generalist agents: 1) an environment that supports a multitude of tasks and goals, 2) a large-scale database of multimodal knowledge, and 3) a flexible and scalable agent architecture. We introduce MineDojo, a new framework built on the popular Minecraft game that features a simulation suite with thousands of diverse open-ended tasks and an internet-scale knowledge base with Minecraft videos, tutorials, wiki pages, and forum discussions. Using MineDojo's data, we propose a novel agent learning algorithm that leverages large pre-trained video-language models as a learned reward function. Our agent is able to solve a variety of open-ended tasks specified in free-form language without any manually designed dense shaping reward. We open-source the simulation suite and knowledge bases (https://minedojo.org) to promote research towards the goal of generally capable embodied agents.
Abstract:Multi-person pose estimation plays an important role in many fields. Although previous works have researched a lot on different parts of human pose estimation, full body pose estimation for multi-person still needs further research. Our work has developed an integrated model through knowledge distillation which can estimate full body poses. Trained based on the AlphaPose system and MSCOCO2017 dataset, our model achieves 51.5 mAP on the validation dataset annotated manually by ourselves. Related resources are available at https://esflfei.github.io/esflfei.gethub.io/website.html.