Fig: An illustration of a motion planning problem, where the robotic car is given the task of inspecting the regions labeled "p1," and "p2."
Mailing Address:| Erion Plaku |
| Rice University |
| Computer Science Dept. |
| 6100 Main St., MS-132 |
| Houston, TX 77005 USA |
| phone: 713-348-2286 |
| fax: 713-348-5930 |
| email: plakue/AT/cs.rice.edu |
Research
A significant challenge confronting autonomous robotics in transportation, exploration of unknown and hazardous environments, search-and-rescue missions, surgery, entertainment and education, and air traffic management lies in the area of automatic motion planning. While progress has been made in planning paths that avoid collisions, it remains particularly challenging to plan motions that enable robots with nonlinear dynamics to complete high-level tasks:
"Some of the most significant challenges confronting autonomous
robotics lie in the area of automatic motion planning. The goal is to
be able to specify a task in a high-level language and have the robot
automatically compile this specification into a set of low-level
motion primitives, or feedback controllers, to accomplish this task."
[Principles of Robot Motion, Choset et al., MIT Press 2005]
The approach I have followed in my Ph.D. research to make progress
toward this fundamental goal in autonomous robotics has been through
the development of novel algorithms and computational
methods that combine research in robotics, hybrid
systems, logic, artificial intelligence (AI),
and machine learning.
The algorithms and methods I have developed in
robotics have made it possible to automatically and efficiently
plan low-level motions that enable increasingly complex robotic and
intelligent systems such as
[Principles of Robot Motion, Choset et al., MIT Press 2005]
- high-dimensional and multi-robot systems (see SRT project)
- systems with nonlinear dynamics such as cars, differential drives, unicycles, tractor trailors, and others (see DSLX project)
- hybrid systems equipped with sophisticated embedded controllers that use discrete logic to modify the underlying dynamics in response to changes in the environment, mishaps, or partial failures (see HyDICE project)
Computational Methods Developed
- SRT: Sampling-based Roadmap of Trees A Platform for High-Dimensional and Multi-Robot Motion Planning
- DSLX: Discrete Search Leading continuous eXploration A Novel Approach to Kinodynamic Motion Planning
- HyDICE: Hybrid Discrete Continuous Exploration Motion Planning and Discovery of Safety Violations for Hybrid Systems
- hcDPES: Hill-Climbing Distance-based Projection onto Euclidean Space Improving Motion Planning and Nonlinear Dimensionality Reduction by Computing Proximity Relations (Nearest Neighbors) Faster
- DPES-ScIMAP: Dimensionality Reduction in Computational Biology Fast and Reliable Analysis of Molecular Motions
- DSRT: Distributed Sampling-based Roadmap of Trees A Scalable and Distributed Platform for High-Dimensional Motion Planning
- DKNNG: Distributed k-Nearest Neighbors Graph A Scalable and Distributed Platform for Nearest-Neighbors Computations from Extensively Large Data Sets
- OOPSMP: Object-Oriented Programming System for Motion Planning A plug-and-play package for motion-planning research or teaching robotics