It is inherently ambiguous to lift 2D results from pre-trained diffusion models to a 3D world for text-to-3D generation. 2D diffusion models solely learn view-agnostic priors and thus lack 3D knowledge during the lifting, leading to the multi-view inconsistency problem. We find that this problem primarily stems from geometric inconsistency, and avoiding misplaced geometric structures substantially mitigates the problem in the final outputs. Therefore, we improve the consistency by aligning the 2D geometric priors in diffusion models with well-defined 3D shapes during the lifting, addressing the vast majority of the problem. This is achieved by fine-tuning the 2D diffusion model to be viewpoint-aware and to produce view-specific coordinate maps of canonically oriented 3D objects. In our process, only coarse 3D information is used for aligning. This "coarse" alignment not only resolves the multi-view inconsistency in geometries but also retains the ability in 2D diffusion models to generate detailed and diversified high-quality objects unseen in the 3D datasets. Furthermore, our aligned geometric priors (AGP) are generic and can be seamlessly integrated into various state-of-the-art pipelines, obtaining high generalizability in terms of unseen shapes and visual appearance while greatly alleviating the multi-view inconsistency problem. Our method represents a new state-of-the-art performance with an 85+% consistency rate by human evaluation, while many previous methods are around 30%. Our project page is https://sweetdreamer3d.github.io/
Transparent objects are common in daily life. However, depth sensing for transparent objects remains a challenging problem. While learning-based methods can leverage shape priors to improve the sensing quality, the labor-intensive data collection in the real world and the sim-to-real domain gap restrict these methods' scalability. In this paper, we propose a method to finetune a stereo network with sparse depth labels automatically collected using a probing system with tactile feedback. We present a novel utility function to evaluate the benefit of touches. By approximating and optimizing the utility function, we can optimize the probing locations given a fixed touching budget to better improve the network's performance on real objects. We further combine tactile depth supervision with a confidence-based regularization to prevent over-fitting during finetuning. To evaluate the effectiveness of our method, we construct a real-world dataset including both diffuse and transparent objects. Experimental results on this dataset show that our method can significantly improve real-world depth sensing accuracy, especially for transparent objects.
Deep reinforcement learning (RL) excels in various control tasks, yet the absence of safety guarantees hampers its real-world applicability. In particular, explorations during learning usually results in safety violations, while the RL agent learns from those mistakes. On the other hand, safe control techniques ensure persistent safety satisfaction but demand strong priors on system dynamics, which is usually hard to obtain in practice. To address these problems, we present Safe Set Guided State-wise Constrained Policy Optimization (S-3PO), a pioneering algorithm generating state-wise safe optimal policies with zero training violations, i.e., learning without mistakes. S-3PO first employs a safety-oriented monitor with black-box dynamics to ensure safe exploration. It then enforces a unique cost for the RL agent to converge to optimal behaviors within safety constraints. S-3PO outperforms existing methods in high-dimensional robotics tasks, managing state-wise constraints with zero training violation. This innovation marks a significant stride towards real-world safe RL deployment.
Negative sampling is essential for implicit-feedback-based collaborative filtering, which is used to constitute negative signals from massive unlabeled data to guide supervised learning. The state-of-the-art idea is to utilize hard negative samples that carry more useful information to form a better decision boundary. To balance efficiency and effectiveness, the vast majority of existing methods follow the two-pass approach, in which the first pass samples a fixed number of unobserved items by a simple static distribution and then the second pass selects the final negative items using a more sophisticated negative sampling strategy. However, selecting negative samples from the original items is inherently restricted, and thus may not be able to contrast positive samples well. In this paper, we confirm this observation via experiments and introduce two limitations of existing solutions: ambiguous trap and information discrimination. Our response to such limitations is to introduce augmented negative samples. This direction renders a substantial technical challenge because constructing unconstrained negative samples may introduce excessive noise that distorts the decision boundary. To this end, we introduce a novel generic augmented negative sampling paradigm and provide a concrete instantiation. First, we disentangle hard and easy factors of negative items. Next, we generate new candidate negative samples by augmenting only the easy factors in a regulated manner: the direction and magnitude of the augmentation are carefully calibrated. Finally, we design an advanced negative sampling strategy to identify the final augmented negative samples, which considers not only the score function used in existing methods but also a new metric called augmentation gain. Extensive experiments on real-world datasets demonstrate that our method significantly outperforms state-of-the-art baselines.
Recently, Neural Radiance Fields (NeRF) has exhibited significant success in novel view synthesis, surface reconstruction, etc. However, since no physical reflection is considered in its rendering pipeline, NeRF mistakes the reflection in the mirror as a separate virtual scene, leading to the inaccurate reconstruction of the mirror and multi-view inconsistent reflections in the mirror. In this paper, we present a novel neural rendering framework, named Mirror-NeRF, which is able to learn accurate geometry and reflection of the mirror and support various scene manipulation applications with mirrors, such as adding new objects or mirrors into the scene and synthesizing the reflections of these new objects in mirrors, controlling mirror roughness, etc. To achieve this goal, we propose a unified radiance field by introducing the reflection probability and tracing rays following the light transport model of Whitted Ray Tracing, and also develop several techniques to facilitate the learning process. Experiments and comparisons on both synthetic and real datasets demonstrate the superiority of our method. The code and supplementary material are available on the project webpage: https://zju3dv.github.io/Mirror-NeRF/.
Non-line-of-sight (NLOS) imaging allows for the imaging of objects around a corner, which enables potential applications in various fields such as autonomous driving, robotic vision, medical imaging, security monitoring, etc. However, the quality of reconstruction is challenged by low signal-noise-ratio (SNR) measurements. In this study, we present a regularization method, referred to as structure sparsity (SS) regularization, for denoising in NLOS reconstruction. By exploiting the prior knowledge of structure sparseness, we incorporate nuclear norm penalization into the cost function of directional light-cone transform (DLCT) model for NLOS imaging system. This incorporation effectively integrates the neighborhood information associated with the directional albedo, thereby facilitating the denoising process. Subsequently, the reconstruction is achieved by optimizing a directional albedo model with SS regularization using fast iterative shrinkage-thresholding algorithm. Notably, the robust reconstruction of occluded objects is observed. Through comprehensive evaluations conducted on both synthetic and experimental datasets, we demonstrate that the proposed approach yields high-quality reconstructions, surpassing the state-of-the-art reconstruction algorithms, especially in scenarios involving short exposure and low SNR measurements.
Following a leading vehicle is a daily but challenging task because it requires adapting to various traffic conditions and the leading vehicle's behaviors. However, the question `Does the following vehicle always actively react to the leading vehicle?' remains open. To seek the answer, we propose a novel metric to quantify the interaction intensity within the car-following pairs. The quantified interaction intensity enables us to recognize interactive and non-interactive car-following scenarios and derive corresponding policies for each scenario. Then, we develop an interaction-aware switching control framework with interactive and non-interactive policies, achieving a human-level car-following performance. The extensive simulations demonstrate that our interaction-aware switching control framework achieves improved control performance and data efficiency compared to the unified control strategies. Moreover, the experimental results reveal that human drivers would not always keep reacting to their leading vehicle but occasionally take safety-critical or intentional actions -- interaction matters but not always.
Reinforcement Learning (RL) algorithms have shown tremendous success in simulation environments, but their application to real-world problems faces significant challenges, with safety being a major concern. In particular, enforcing state-wise constraints is essential for many challenging tasks such as autonomous driving and robot manipulation. However, existing safe RL algorithms under the framework of Constrained Markov Decision Process (CMDP) do not consider state-wise constraints. To address this gap, we propose State-wise Constrained Policy Optimization (SCPO), the first general-purpose policy search algorithm for state-wise constrained reinforcement learning. SCPO provides guarantees for state-wise constraint satisfaction in expectation. In particular, we introduce the framework of Maximum Markov Decision Process, and prove that the worst-case safety violation is bounded under SCPO. We demonstrate the effectiveness of our approach on training neural network policies for extensive robot locomotion tasks, where the agent must satisfy a variety of state-wise safety constraints. Our results show that SCPO significantly outperforms existing methods and can handle state-wise constraints in high-dimensional robotics tasks.
Human-robot collaboration (HRC) is one key component to achieving flexible manufacturing to meet the different needs of customers. However, it is difficult to build intelligent robots that can proactively assist humans in a safe and efficient way due to several challenges.First, it is challenging to achieve efficient collaboration due to diverse human behaviors and data scarcity. Second, it is difficult to ensure interactive safety due to uncertainty in human behaviors. This paper presents an integrated framework for proactive HRC. A robust intention prediction module, which leverages prior task information and human-in-the-loop training, is learned to guide the robot for efficient collaboration. The proposed framework also uses robust safe control to ensure interactive safety under uncertainty. The developed framework is applied to a co-assembly task using a Kinova Gen3 robot. The experiment demonstrates that our solution is robust to environmental changes as well as different human preferences and behaviors. In addition, it improves task efficiency by approximately 15-20%. Moreover, the experiment demonstrates that our solution can guarantee interactive safety during proactive collaboration.