A5037969179- Autonomous Vehicle Technology and Safety
- Robotic Path Planning Algorithms
- Robotics and Sensor-Based Localization
- Remote Sensing and LiDAR Applications
- Advanced Vision and Imaging
- Video Surveillance and Tracking Methods
- Robotic Locomotion and Control
- Planetary Science and Exploration
- Computer Graphics and Visualization Techniques
- Space Exploration and Technology
- 3D Surveying and Cultural Heritage
- Advanced Software Engineering Methodologies
- Prosthetics and Rehabilitation Robotics
- Real-Time Systems Scheduling
- Systems Engineering Methodologies and Applications
- Data Management and Algorithms
- Advanced Neural Network Applications
- Biomedical and Engineering Education
- Astro and Planetary Science
- Teleoperation and Haptic Systems
- Modular Robots and Swarm Intelligence
- Distributed systems and fault tolerance
- Robot Manipulation and Learning
- 3D Shape Modeling and Analysis
- Traffic control and management
Virginia Tech
2015-2022
Carnegie Mellon University
2006-2010
Abstract Boss is an autonomous vehicle that uses on‐board sensors (global positioning system, lasers, radars, and cameras) to track other vehicles, detect static obstacles, localize itself relative a road model. A three‐layer planning system combines mission, behavioral, motion drive in urban environments. The mission layer considers which street take achieve goal. behavioral determines when change lanes precedence at intersections performs error recovery maneuvers. selects actions avoid...
Recent advances in 3D sensing have created unique challenges for computer vision. One fundamental challenge is finding a good representation sensor data. Most popular representations (such as PointNet) are proposed the context of processing truly data (e.g. points sampled from mesh models), ignoring fact that sensored such LiDAR sweep 2.5D. We argue representing 2.5D collections (x,y,z) fundamentally destroys hidden information about freespace. In this paper, we demonstrate knowledge can be...
Abstract This article presents a robust approach to navigating at high speed across desert terrain. A central theme of this is the combination simple ideas and components build capable system. pair robots were developed, which completed 212 km Grand Challenge race in approximately 7 h. pathcentric navigation system uses LIDAR RADAR based perception sensors traverse trails avoid obstacles speeds up 15 m/s. The onboard leverages human‐based preplanning improve reliability robustness. have been...
The Urban Challenge represents a technological leap beyond the previous Grand Challenges. challenge encompasses three primary behaviors: driving on roads, handling intersections and maneuvering in zones. In implementing urban we have decomposed problem into five components. Mission Planning determines an efficient route through network of roads. A behavioral layer executes environment, adapting to local traffic exceptional situations as necessary. motion planning safeguards robot by...
The task of teleoperating a robot over wireless video link is known to be very difficult. Teleoperation becomes even more difficult when the surrounded by dense obstacles, or speed requirements are high, quality poor, links subject latency. Due high-quality lidar data, and improvements in computing compression, virtualized reality has capacity dramatically improve teleoperation performance — high-speed situations that were formerly impossible. In this paper, we demonstrate conversion...
This paper describes an algorithm for autonomous car to identify the shape of a roadway by detecting geometric features via LIDAR. The data from multiple LIDAR are fused together detect both obstacles as well such curbs, berms, and shoulders. These boundaries used stochastic state estimator most likely road shape. has been successfully allow drive on paved roadways off-road trails without requiring different sets parameters domains.
This paper describes a method for creating photorealistic three-dimensional (3D) models of real-world environments in real-time the purpose improving and extending capabilities vehicle tele-operation. Our approach utilizes combined data from laser scanner (for modeling 3D geometry) video camera surface appearance). The sensors are mounted on moving platform, photo-realistic model vehicle's environment is generated displayed to remote operator real time. consists three main components:...
The Electric Series Compliant Humanoid for Emergency Response (ESCHER) platform represents the culmination of four years development at Virginia Tech to produce a full‐sized force‐controlled humanoid robot capable operating in unstructured environments. ESCHER's locomotion capability was demonstrated DARPA Robotics Challenge (DRC) Finals when it successfully navigated 61 m loose dirt course. Team VALOR, Track A team, developed ESCHER leveraging and improving upon bipedal technologies...
The DARPA Urban Challenge required robots to drive 60 miles on suburban roads while following the rules of road in interactions with human drivers and other robots. Tartan Racing’s Boss won competition, completing course just over 4 hours. This paper describes software infrastructure developed by team support perception, planning, behavior generation, artificial intelligence components Boss. We discuss organizing principles infrastructure, as well details operator interface, interprocess...
This paper presents a fast global scan matching technique for high-speed vehicle navigation. The proposed grid-based scan-to-map collectively handles unprocessed points at each grid cell as feature. features are transformed and located in the frame updated every time new is acquired. Since registered only features, which mean of cell, feature very fast. Representation by multiple further maintains accuracy regardless size while processing achieved. therefore suited localization Experimental...
State-of-the-art lidar panoptic segmentation (LPS) methods follow "bottom-up" segmentation-centric fashion wherein they build upon semantic networks by utilizing clustering to obtain object instances. In this paper, we re-think approach and propose a surprisingly simple yet effective detection-centric network for both LPS tracking. Our is modular design optimized all aspects of the tracking task. One core components our instance detection branch, which train using point-level (modal)...
Recent advances in 3D sensing have created unique challenges for computer vision. One fundamental challenge is finding a good representation sensor data. Most popular representations (such as PointNet) are proposed the context of processing truly data (e.g. points sampled from mesh models), ignoring fact that sensored such LiDAR sweep 2.5D. We argue representing 2.5D collections (x, y, z) fundamentally destroys hidden information about freespace. In this paper, we demonstrate knowledge can...
The design and organization of complex robotic systems traditionally requires laborious trial-and-error processes to ensure both hardware software components are correctly connected with the resources necessary for computation. This paper presents a novel generalization quadratic assignment routing problem, introducing formalisms selecting interconnections synthesize complete system capable providing some user-defined functionality. By mission context, functional requirements, modularity...
State-of-the-art lidar panoptic segmentation (LPS) methods follow bottom-up segmentation-centric fashion wherein they build upon semantic networks by utilizing clustering to obtain object instances. In this paper, we re-think approach and propose a surprisingly simple yet effective detection-centric network for both LPS tracking. Our is modular design optimized all aspects of the tracking task. One core components our instance detection branch, which train using point-level (modal)...