The Ben Franklin Racing Team’s goal is to build fast, reliable, safe and autonomous vehicles that will revolutionize transportation systems in urban environments. We will leverage state-of-the-art advances in sensing, control theory, machine learning, automotive technology and artificial advantages to build robotic cars. The team will participate the 2007 DARPA Urban Challenge.

This multi-university project involves the University of Pennsylvania, the University of California at Berkeley, and Carnegie Mellon University. It focuses on the design and evaluation of the adaptive hierarchical control of mixed autonomous and human operated semi-autonomous teams that deliver high levels of mission reliability despite uncertainty arising from rapidly evolving environments and malicious interference from an intelligent adversary. Equipment for this project is supported by an ARO DURIP grant.

This multi-university project brings together experts in artificial intelligence, control theory, robotics, systems engineering and biology with the goal of understanding swarming behaviors in nature and applications of biologically-inspired models of swarm behaviors to large networked groups of autonomous vehicles. This multi-university project is led by the University of Pennsylvania and will be performed in collaboration with the Massachusetts Institute of Technology, the University of California at Berkeley, the University of California at Santa Barbara, and Yale University.

This project is investigating and developing methods for the recovery of 3D underground structures from subsurface non-invasive measurements obtained with ground penetrating radar, magnetometry, and conductivity sensors. The results will not only provide hints for further excavation but also 3D models that can be studied as if they were already excavated. The three fundamental challenges investigated are the inverse problem of recovering the volumetric material distribution, the segmentation of the underground volumes, and the reconstruction of the surfaces that comprise interesting structures.

The goal of the LAGR program is to develop a new generation of learned perception and control algorithms for autonomous ground vehicles, and to integrate these learned algorithms with a highly capable robotic ground vehicle.

This project, in collaboration with the University of Minnesota and the California Institute of Technology, addresses research issues key to an important application of robot teams and information technology (emergency response in hazardous environments for various tasks). The research focuses on the development of methods for team coordination and dynamic distribution of tasks to robots. The project integrates the algorithms with first responder teams, emphasizing realistic scenarios.

Aims to develop a modular robot that consists of many reconfigurable modules and demonstrate its multifunction and reconfiguration in a desert for running, climbing, structuring, life-protecting, and flying. We have built a first generation module with a single degree of freedom and multiple connection ports on different faces.

HURT is a multi-vehicle controller that coordinates and collaboratively plans urban RSTA missions for autonomous vehicles.  It implements augmented autonomy for teams of arbitrary vehicle platforms.

We present a general graph learning algorithm for spectral graph partitioning, that allows direct supervised learning of graph structures. Learning is based on gradient descent in the space of graph weights, using derivatives of eigenvectors. This algorithm effectively learns a graph capable of memorizing and retrieving multiple patterns given noisy inputs. We experimented on segmentation and recognition tasks, including  bottom-up geometric shape extraction with top-down priors, and hand-written digit recognition.

This project goal is to design, develop, and implement several new algorithms and architectures for learning controllers for high-speed quadruped locomotion over rough terrain. This will be achived by incorporating a dynamically relevant lowdimensional representation of the joint trajectories for control and learning. The low-dimensional space of control parameters will be automatically learned from examples of high - dimensional joint trajectories, and these parameters will be used to compactly describe a number of primitive gaitmotions. Using a formal compositional semantics, the primitive gaits will be temporally sequenced in a hierarchical manner to generatemore complex locomotionmanuevers. Reinforcement learning techniques will be applied to optimize the switching boundaries between these primitive locomotionmodes, as well as tune the underlying low-dimensional controlparameters for speed and robustness.

We present a multiscale graph-based image segmentation algorithm. In contrast to most multiscale image processing, this algorithm works on multiple scales of the image in parallel, without iteration, to capture both coarse and fine level details. We demonstrate that large image segmentation graphs can be compressed into multiple scales capturing image structure at increasingly large neighborhood. The algorithm has O(N) time complexity, allowing to segment large images with typically N = 1000 x 1000 pixels.

We consider the problem of recovering an underwater image distorted by surface waves. Our experimental setup consists of a camera positioned above a swimming pool facing down and a book lying on the bottom of the pool. A large amount of video data of the distorted image, e.g. the cover of a book, is acquired and the problem is posed in terms of understanding the statistics of local patches in the image plane. This challenging reconstruction task can be formulated as a manifold learning problem, such that the center of the manifold is the image of the undistorted patch. To compute the center, we present a new technique to estimate global distances on the manifold.

The BIOCOMP project applies hybrid systems to modeling and simulation of metabolic and cellular control pathways. Hybrid systems combine both discrete events and continuous differential equations, unlike traditional approaches choosing exclusively between discrete or continuous dynamics. These models capture the switching behavior in phenomena such as transcription, protein-protein interactions, and cell division and growth.

The DaVinci project brings together mathematicians and engineers to study systems that can be modeled by Differential Algebraic Inequalities and Differential Complementarity Problems. The goal is to develop a set of mathematical and computational tools broadly applicable to multiple engineering disciplines, including robotics, manufacturing, chemical processes, hydraulic processes, avionics, intelligent highways, and automotive systems.

This project develops algorithms to recognize human activity from unsupervised video streams. Detection and classification address multiple levels of abstraction, including limb tracking, human identification, gesture recognition, and activity inference. The ultimate goal is to develop computational algorithms to understand human behavior in video.

Control and decision-making for independent legged robotic agents.

This research develops methodology and software for deploying multiple autonomous robots in an unstructured and unknown environment. Its framework of supervised autonomy enables both deliberate and reactive behavior for the robots during autonomous operation as they adapt to their environment and learn new tasks. It also permits a human to dynamically reprogram the robots by teleoperation. Applications span reconnaissance, surveillance, target acquisition, and removal of explosive ordnance.

In this work we employ epipolar plane image analysis to recover the positions of edge features in the scene. Once we have recovered the positions of these salient points we can use a morphing technique to synthesize new views of the scene.

Omnidirectional vision systems can provide panoramic alertness in surveillance, improve navigational capabilities, and produce panoramic images for multimedia.

This project is an effort at the GRASP Laboratory to develop a new technology in the form of a smart wheelchair. This device is equipped with a virtual interface and on-board cameras that enable the subject to navigate on the ground by interacting with the virtual system interface or use one of the built-in control algorithms.

This project dealt with the problem of recovering models of articulated figures, including humans, from single snapshots acquired with an uncalibrated camera. The resulting reconstruction algorithm can be used to recover stick figure models from newspaper photos or web site photos. It has also been used to recover models of moving figures from short video sequences.

Tele-Immersion will enable users at geographically distributed sites to collaborate in real time in a shared, environment as if they were in the same physical room. This new paradigm for human-computer interaction is the ultimate synthesis of networking and media technologies.

The main motivation for the project is to develop cooperative behavior for between unmanned aerial vehicles and or ground vehicles at the GRASP Lab. Another motivation is to develop control algorithms methodologies to allow the aircraft to form a part of a heterogeneous robot team including ground and other aerial vehicles and perform mission tasks at higher levels.

A method for estimating the trajectory of a moving camera and the appearance of a scene from omnidirectional video sequences has been developed. The end result of our procedure is an omnidirectional video sequence where each frame is augmented with pose information and a sparse 3D model of the scene.