arXiv subjects

Mathew Schwartz

Publications and source records attributed to Mathew Schwartz.

11 recordsLinked to original sources

arXiv2026

JACoP: Joint Alignment for Compliant Multi-Agent Prediction

Stochastic Human Trajectory Prediction (HTP) using generative modeling has emerged as a significant area of research. Although state-of-the-art models excel in optimizing the accuracy of individual agents, they often struggle to generate predictions that are collectively compliant, leading to output trajectories marred by social collisions and environmental violations, thus rendering them impractical for real-world applications. To bridge this gap, we present JACoP: Joint Alignment for Compliant Multi-Agent Prediction, an innovative multi-stage framework that ensures scene-level plausibility. JACoP incorporates an Anchor-Based Agent-Centric Profiler for effective initial compliance filtering and employs a Markov Random Field (MRF) based aligner to formalize the joint selection for scene predictions. By representing inter-agent spatial and social costs as MRF energy potentials, we successfully infer and sample from the joint trajectory distribution, achieving prediction with optimal scene compliance. Comprehensive experiments show that JACoP not only achieves competitive accuracy, but also sets a new standard in reducing both environmental violations and social collisions, thereby confirming its ability to produce collectively feasible and practically applicable trajectory predictions.

cs.CV↗

arXiv2026

ECTraj: Enhanced Consistency Training for Multi-Agent Trajectory Prediction

Diffusion models for multi-agent trajectory prediction are limited by iterative denoising, which causes inference latency that hinders their use in time-critical settings like autonomous driving. Fast-sampling variants using DDIM and informed initial noise distributions partially alleviate this issue, but they either fail to achieve true single-step generation or are constrained by the chosen noise distribution. Consistency Models (CMs) offer high-quality one-step generation by mapping noise directly to data, but are difficult to train from scratch. We propose ECTraj, an enhanced CM pipeline with improved training and conditional generation for trajectory prediction. Our framework extends the student-teacher consistency training scheme: the student produces standard outputs, while the teacher explicitly fuses its predictions with parts of the ground truth to give stronger supervision. We also exploit CMs' direct denoising for top-K multi-shot generation during training. Combining conditional generation with this enhanced consistency objective yields faster inference and improved prediction accuracy, establishing competitive new benchmarks on the large-scale Argoverse 2 dataset.

cs.CV↗

arXiv2024

Microscopic modeling of attention-based movement behaviors

For transportation hubs, leveraging pedestrian flows for commercial activities presents an effective strategy for funding maintenance and infrastructure improvements. However, this introduces new challenges, as consumer behaviors can disrupt pedestrian flow and efficiency. To optimize both retail potential and pedestrian efficiency, careful strategic planning in store layout and facility dimensions was done by expert judgement due to the complexity in pedestrian dynamics in the retail areas of transportation hubs. This paper introduces an attention-based movement model to simulate these dynamics. By simulating retail potential of an area through the duration of visual attention it receives, and pedestrian efficiency via speed loss in pedestrian walking behaviors, the study further explores how design features can influence the retail potential and pedestrian efficiency in a bi-directional corridor inside a transportation hub. Project webpage: https://danruili.github.io/AttentionMove

physics.soc-ph↗

arXiv2023

Torque-based Deep Reinforcement Learning for Task-and-Robot Agnostic Learning on Bipedal Robots Using Sim-to-Real Transfer

In this paper, we review the question of which action space is best suited for controlling a real biped robot in combination with Sim2Real training. Position control has been popular as it has been shown to be more sample efficient and intuitive to combine with other planning algorithms. However, for position control gain tuning is required to achieve the best possible policy performance. We show that instead, using a torque-based action space enables task-and-robot agnostic learning with less parameter tuning and mitigates the sim-to-reality gap by taking advantage of torque control's inherent compliance. Also, we accelerate the torque-based-policy training process by pre-training the policy to remain upright by compensating for gravity. The paper showcases the first successful sim-to-real transfer of a torque-based deep reinforcement learning policy on a real human-sized biped robot. The video is available at https://youtu.be/CR6pTS39VRE.

cs.RO↗

arXiv2022

NODE IK: Solving Inverse Kinematics with Neural Ordinary Differential Equations for Path Planning

This paper proposes a novel inverse kinematics (IK) solver of articulated robotic systems for path planning. IK is a traditional but essential problem for robot manipulation. Recently, data-driven methods have been proposed to quickly solve the IK for path planning. These methods can handle a large amount of IK requests at once with the advantage of GPUs. However, the accuracy is still low, and the model requires considerable time for training. Therefore, we propose an IK solver that improves accuracy and memory efficiency by utilizing the continuous hidden dynamics of Neural ODE. The performance is compared using multiple robots.

cs.RO↗

arXiv2022

Optimizing Indoor Navigation Policies For Spatial Distancing

In this paper, we focus on the modification of policies that can lead to movement patterns and directional guidance of occupants, which are represented as agents in a 3D simulation engine. We demonstrate an optimization method that improves a spatial distancing metric by modifying the navigation graph by introducing a measure of spatial distancing of agents as a function of agent density (i.e., occupancy). Our optimization framework utilizes such metrics as the target function, using a hybrid approach of combining genetic algorithm and simulated annealing. We show that within our framework, the simulation-optimization process can help to improve spatial distancing between agents by optimizing the navigation policies for a given indoor environment.

cs.MA↗

arXiv2021

Adding Visibility to Visibility Graphs: Weighting Visibility Analysis with Attenuation Coefficients

Evaluating the built environment based on visibility has been long used as a tool for human-centric design. The origins of isovists and visibility graphs are within interior spaces, while more recently, these evaluation techniques have been applied in the urban context. One of the key differentiators of an outside environment is the weather, which has largely been ignored in the design computation and space-syntax research areas. While a visibility graph is a straightforward metric for determining connectivity between regions of space through a line of sight calculation, this approach largely ignores the actual visibility of one point to another. This paper introduces a new method for weighting a visibility graph based on weather conditions (i.e. rain, fog, snow). These new factors are integrated into visibility graphs and applied to sample environments to demonstrate the variance between assuming a straight line of sight and reduced visibility.

cs.CV↗

arXiv2021

Style Similarity as Feedback for Product Design

Matching and recommending products is beneficial for both customers and companies. With the rapid increase in home goods e-commerce, there is an increasing demand for quantitative methods for providing such recommendations for millions of products. This approach is facilitated largely by online stores such as Amazon and Wayfair, in which the goal is to maximize overall sales. Instead of focusing on overall sales, we take a product design perspective, by employing big-data analysis for determining the design qualities of a highly recommended product. Specifically, we focus on the visual style compatibility of such products. We build off previous work which implemented a style-based similarity metric for thousands of furniture products. Using analysis and visualization, we extract attributes of furniture products that are highly compatible style-wise. We propose a designer in-the-loop workflow that mirrors methods of displaying similar products to consumers browsing e-commerce websites. Our findings are useful when designing new products, since they provide insight regarding what furniture will be strongly compatible across multiple styles, and hence, more likely to be recommended.

cs.CV↗

arXiv2021

Linking Physical Objects to Their Digital Twins via Fiducial Markers Designed for Invisibility to Humans

The ability to label and track physical objects that are assets in digital representations of the world is foundational to many complex systems. Simple, yet powerful methods such as bar- and QR-codes have been highly successful, e.g. in the retail space, but the lack of security, limited information content and impossibility of seamless integration with the environment have prevented a large-scale linking of physical objects to their digital twins. This paper proposes to link digital assets created through BIM with their physical counterparts using fiducial markers with patterns defined by Cholesteric Spherical Reflectors (CSRs), selective retroreflectors produced using liquid crystal self-assembly. The markers leverage the ability of CSRs to encode information that is easily detected and read with computer vision while remaining practically invisible to the human eye. We analyze the potential of a CSR-based infrastructure from the perspective of BIM, critically reviewing the outstanding challenges in applying this new class of functional materials, and we discuss extended opportunities arising in assisting autonomous mobile robots to reliably navigate human-populated environments, as well as in augmented reality.

cond-mat.soft↗

arXiv2021

Embedding Intelligence in Materials for Responsive Built Environment: A Topical Review on Liquid Crystal Elastomer Actuators and Sensors

Liquid Crystal Elastomers (LCEs) are an exciting category of material that has tremendous application potential across a variety of fields, owing to their unique properties that enable both sensing and actuation. To some, LCEs are simply another type of Shape Memory Polymer, while to others they are an interesting on-going scientific experiment. In this visionary article, we bring an interdisciplinary discussion around creative and impactful ways that LCEs can be applied in the Built Environment to support kinematic and kinetic buildings and situational awareness. We focus particularly on the autonomy made possible by using LCEs, potentially removing needs for motors, wiring and tubing, and even enabling fully independent operation in response to natural environment variations, requiring no power sources. To illustrate the potential, we propose a number of concrete application scenarios where LCEs could offer innovative solutions to problems of great societal importance, such as autonomous active ventilation, heliotropic solar panel systems which can also remove snow or sand autonomously, and invisible coatings with strain mapping functionality, alerting residents in case of dangerous (static or dynamic) loads on roofs or windows, as well as assisting building safety inspection teams after earthquakes.

cond-mat.soft↗

arXiv2021

Human Centric Accessibility Graph For Environment Analysis

Understanding design decisions in relation to the future occupants of a building is a crucial part of good design. However, limitations in tools and expertise hinder meaningful human-centric decisions during the design process. In this paper, a novel Spatial Human Accessibility graph for Planning and Environment Analysis (SHAPE) is introduced that brings together the technical challenges of discrete representations of digital models, with human-based metrics for evaluating the environment. SHAPE: does not need labeled geometry as input, works with multi-level buildings, captures surface variations (e.g., slopes in a terrain), and can be used with existing graph theory (e.g., gravity, centrality) techniques. SHAPE uses ray-casting to perform a search, generating a dense graph of all accessible locations within the environment and storing the type of travel required in a graph (e.g., up a slope, down a step). The ability to simultaneously evaluate and plan paths from multiple human factors is shown to work on digital models across room, building, and topography scales. The results enable designers and planners to evaluate options of the built environment in new ways, and at higher fidelity, that will lead to more human-friendly and accessible environments.

cs.GR↗