UnrealZoo: Enriching Photo-realistic Virtual Worlds for Embodied AI

We introduce UnrealZoo, a rich collection of photo-realistic 3D virtual worlds built on Unreal Engine, designed to reflect the complexity and variability of the open worlds. Additionally, we offer a variety of playable entities for embodied AI agents. Based on UnrealCV, we provide a suite of easy-to-use Python APIs and tools for various potential applications, such as data collection, environment augmentation, distributed training, and benchmarking. We optimize the rendering and communication efficiency of UnrealCV to support advanced applications, such as multi-agent interaction. Our experiments benchmark agents in various complex scenes, focusing on visual navigation and tracking, which are fundamental capabilities for embodied visual intelligence. The results yield valuable insights into the advantages of diverse training environments for reinforcement learning (RL) agents and the challenges faced by current embodied vision agents, including those based on RL and large vision-language models (VLMs), in open worlds. These challenges involve latency in closed-loop control in dynamic scenes and reasoning about 3D spatial structures in unstructured terrain.

SDRS: Shape-Differentiable Robot Simulator

Robot simulators are indispensable tools across many fields, and recent research has significantly improved their functionality by incorporating additional gradient information. However, existing differentiable robot simulators suffer from non-differentiable singularities, when robots undergo substantial shape changes. To address this, we present the Shape-Differentiable Robot Simulator (SDRS), designed to be differentiable under significant robot shape changes. The core innovation of SDRS lies in its representation of robot shapes using a set of convex polyhedrons. This approach allows us to generalize smooth, penalty-based contact mechanics for interactions between any pair of convex polyhedrons. Using the separating hyperplane theorem, SDRS introduces a separating plane for each pair of contacting convex polyhedrons. This separating plane functions as a zero-mass auxiliary entity, with its state determined by the principle of least action. This setup ensures global differentiability, even as robot shapes undergo significant geometric and topological changes. To demonstrate the practical value of SDRS, we provide examples of robot co-design scenarios, where both robot shapes and control movements are optimized simultaneously.

XFormParser: A Simple and Effective Multimodal Multilingual Semi-structured Form Parser

In the domain of document AI, semi-structured form parsing plays a crucial role. This task leverages techniques from key information extraction (KIE), dealing with inputs that range from plain text to intricate modal data comprising images and structural layouts. The advent of pre-trained multimodal models has driven the extraction of key information from form documents in different formats such as PDFs and images. Nonetheless, the endeavor of form parsing is still encumbered by notable challenges like subpar capabilities in multi-lingual parsing and diminished recall in contexts rich in text and visuals. In this work, we introduce a simple but effective \textbf{M}ultimodal and \textbf{M}ultilingual semi-structured \textbf{FORM} \textbf{PARSER} (\textbf{XFormParser}), which is anchored on a comprehensive pre-trained language model and innovatively amalgamates semantic entity recognition (SER) and relation extraction (RE) into a unified framework, enhanced by a novel staged warm-up training approach that employs soft labels to significantly refine form parsing accuracy without amplifying inference overhead. Furthermore, we have developed a groundbreaking benchmark dataset, named InDFormBench, catering specifically to the parsing requirements of multilingual forms in various industrial contexts. Through rigorous testing on established multilingual benchmarks and InDFormBench, XFormParser has demonstrated its unparalleled efficacy, notably surpassing the state-of-the-art (SOTA) models in RE tasks within language-specific setups by achieving an F1 score improvement of up to 1.79\%. Our framework exhibits exceptionally improved performance across tasks in both multi-language and zero-shot contexts when compared to existing SOTA benchmarks. The code is publicly available at https://github.com/zhbuaa0/layoutlmft.

BehaviorGPT: Smart Agent Simulation for Autonomous Driving with Next-Patch Prediction

Simulating realistic interactions among traffic agents is crucial for efficiently validating the safety of autonomous driving systems. Existing leading simulators primarily use an encoder-decoder structure to encode the historical trajectories for future simulation. However, such a paradigm complicates the model architecture, and the manual separation of history and future trajectories leads to low data utilization. To address these challenges, we propose Behavior Generative Pre-trained Transformers (BehaviorGPT), a decoder-only, autoregressive architecture designed to simulate the sequential motion of multiple agents. Crucially, our approach discards the traditional separation between "history" and "future," treating each time step as the "current" one, resulting in a simpler, more parameter- and data-efficient design that scales seamlessly with data and computation. Additionally, we introduce the Next-Patch Prediction Paradigm (NP3), which enables models to reason at the patch level of trajectories and capture long-range spatial-temporal interactions. BehaviorGPT ranks first across several metrics on the Waymo Sim Agents Benchmark, demonstrating its exceptional performance in multi-agent and agent-map interactions. We outperformed state-of-the-art models with a realism score of 0.741 and improved the minADE metric to 1.540, with an approximately 91.6% reduction in model parameters.

Empowering Embodied Visual Tracking with Visual Foundation Models and Offline RL

Embodied visual tracking is to follow a target object in dynamic 3D environments using an agent's egocentric vision. This is a vital and challenging skill for embodied agents. However, existing methods suffer from inefficient training and poor generalization. In this paper, we propose a novel framework that combines visual foundation models (VFM) and offline reinforcement learning (offline RL) to empower embodied visual tracking. We use a pre-trained VFM, such as ``Tracking Anything", to extract semantic segmentation masks with text prompts. We then train a recurrent policy network with offline RL, e.g., Conservative Q-Learning, to learn from the collected demonstrations without online agent-environment interactions. To further improve the robustness and generalization of the policy network, we also introduce a mask re-targeting mechanism and a multi-level data collection strategy. In this way, we can train a robust tracker within an hour on a consumer-level GPU, e.g., Nvidia RTX 3090. Such efficiency is unprecedented for RL-based visual tracking methods. We evaluate our tracker on several high-fidelity environments with challenging situations, such as distraction and occlusion. The results show that our agent outperforms state-of-the-art methods in terms of sample efficiency, robustness to distractors, and generalization to unseen scenarios and targets. We also demonstrate the transferability of the learned tracker from the virtual world to real-world scenarios.

Learning Neural Traffic Rules

Extensive research has been devoted to the field of multi-agent navigation. Recently, there has been remarkable progress attributed to the emergence of learning-based techniques with substantially elevated intelligence and realism. Nonetheless, prevailing learned models face limitations in terms of scalability and effectiveness, primarily due to their agent-centric nature, i.e., the learned neural policy is individually deployed on each agent. Inspired by the efficiency observed in real-world traffic networks, we present an environment-centric navigation policy. Our method learns a set of traffic rules to coordinate a vast group of unintelligent agents that possess only basic collision-avoidance capabilities. Our method segments the environment into distinct blocks and parameterizes the traffic rule using a Graph Recurrent Neural Network (GRNN) over the block network. Each GRNN node is trained to modulate the velocities of agents as they traverse through. Using either Imitation Learning (IL) or Reinforcement Learning (RL) schemes, we demonstrate the efficacy of our neural traffic rules in resolving agent congestion, closely resembling real-world traffic regulations. Our method handles up to $240$ agents at real-time and generalizes across diverse agent and environment configurations.

Second-Order Convergent Collision-Constrained Optimization-Based Planner

Finding robot poses and trajectories represents a foundational aspect of robot motion planning. Despite decades of research, efficiently and robustly addressing these challenges is still difficult. Existing approaches are often plagued by various limitations, such as intricate geometric approximations, violations of collision constraints, or slow first-order convergence. In this paper, we introduce two novel optimization formulations that offer provable robustness, achieving second-order convergence while requiring only a convex approximation of the robot's links and obstacles. Our first method, known as the Explicit Collision Barrier (ECB) method, employs a barrier function to guarantee separation between convex objects. ECB uses an efficient matrix factorization technique, enabling a second-order Newton's method with an iterative complexity linear in the number of separating planes. Our second method, referred to as the Implicit Collision Barrier (ICB) method, further transforms the separating planes into implicit functions of robot poses. We show such an implicit objective function is twice-differentiable, with derivatives evaluated at a linear complexity. To assess the effectiveness of our approaches, we conduct a comparative study with a first-order baseline algorithm across six testing scenarios. Our results unequivocally justify that our method exhibits significantly faster convergence rates compared to the baseline algorithm.

Learning Reduced-Order Soft Robot Controller

Deformable robots are notoriously difficult to model or control due to its high-dimensional configuration spaces. Direct trajectory optimization suffers from the curse-of-dimensionality and incurs a high computational cost, while learning-based controller optimization methods are sensitive to hyper-parameter tuning. To overcome these limitations, we hypothesize that high fidelity soft robots can be both simulated and controlled by restricting to low-dimensional spaces. Under such assumption, we propose a two-stage algorithm to identify such simulation- and control-spaces. Our method first identifies the so-called simulation-space that captures the salient deformation modes, to which the robot's governing equation is restricted. We then identify the control-space, to which control signals are restricted. We propose a multi-fidelity Riemannian Bayesian bilevel optimization to identify task-specific control spaces. We show that the dimension of control-space can be less than $10$ for a high-DOF soft robot to accomplish walking and swimming tasks, allowing low-dimensional MPC controllers to be applied to soft robots with tractable computational complexity.

Learning based 2D Irregular Shape Packing

2D irregular shape packing is a necessary step to arrange UV patches of a 3D model within a texture atlas for memory-efficient appearance rendering in computer graphics. Being a joint, combinatorial decision-making problem involving all patch positions and orientations, this problem has well-known NP-hard complexity. Prior solutions either assume a heuristic packing order or modify the upstream mesh cut and UV mapping to simplify the problem, which either limits the packing ratio or incurs robustness or generality issues. Instead, we introduce a learning-assisted 2D irregular shape packing method that achieves a high packing quality with minimal requirements from the input. Our method iteratively selects and groups subsets of UV patches into near-rectangular super patches, essentially reducing the problem to bin-packing, based on which a joint optimization is employed to further improve the packing ratio. In order to efficiently deal with large problem instances with hundreds of patches, we train deep neural policies to predict nearly rectangular patch subsets and determine their relative poses, leading to linear time scaling with the number of patches. We demonstrate the effectiveness of our method on three datasets for UV packing, where our method achieves a higher packing ratio over several widely used baselines with competitive computational speed.

Killing Two Birds with One Stone: Quantization Achieves Privacy in Distributed Learning

Communication efficiency and privacy protection are two critical issues in distributed machine learning. Existing methods tackle these two issues separately and may have a high implementation complexity that constrains their application in a resource-limited environment. We propose a comprehensive quantization-based solution that could simultaneously achieve communication efficiency and privacy protection, providing new insights into the correlated nature of communication and privacy. Specifically, we demonstrate the effectiveness of our proposed solutions in the distributed stochastic gradient descent (SGD) framework by adding binomial noise to the uniformly quantized gradients to reach the desired differential privacy level but with a minor sacrifice in communication efficiency. We theoretically capture the new trade-offs between communication, privacy, and learning performance.