Maxim Likhachev

I am a Research Assistant Professor with the Department of Computer and Information Science, which is part of School of Engineering and Applied Science, University of Pennsylvania. I am also a member of the GRASP laboratory.

email: myusername@seas.upenn.edu, where myusername is the word right after tilde in the URL of this website
office #: GRW 376
address: 3330 Walnut St, Levine Hall, Philadelphia, PA 19104

Research Statement

Research: My general research interests lie in Artificial Intelligence and Robotics. More specifically, they currently cover planning in deterministic and probabilistic domains and machine learning. So far, my research has been mainly motivated by the problem of fast and intelligent decision making by autonomous robotic systems in real-world environments. I do get easily motivated, however, by other interesting problems in AI.

During my 2-year Postdoctoral Fellowship at CMU, I worked with Tony Stentz on multi-agent planning with adversaries. During the same time, I have also worked on the design and implementation of a complex maneuvering planner for the CMU vehicle that WON the Urban Challenge race (the third DARPA Grand Challenge). During my Ph.D. studies at CMU, my advisors were Geoff Gordon and Sebastian Thrun (presently at Stanford). Before enrolling into the Ph.D. program at CMU I have been a graduate student for two years in the College of Computing at Georgia Tech where I worked with Ronald Arkin at Mobile Robot Laboratory and Sven Koenig.

SELECTED RESEARCH PROJECTS

I research algorithms that solve large-scale real world planning and decision-making problems with uncertainty and real-time performance requirements. I strive to equip these algorithms with a solid theoretical analysis and use them to provide efficient and robust planning in real robotic systems including mobile robots navigating at high-speeds in partially-known environments, multi-robot systems operating in hostile environments and autonomous vehicles navigating urban environments such as the CMU vehicle that won the 1st place in the Urban Challenge race in 2007 (the 3rd DARPA Grand Challenge competition).

My current research thrust is mostly in developing novel graph search techniques that can solve complex deterministic and probabilistic planning problems in robotics. Such algorithms are typically quite general, come with rigorous theoretical analysis and are simple-to-implement. Below are some of the research problems I am currently concentrating on. Some of the developed algorithms are part of the publicly available Search-based Planning Library (SBPL).

Autonomouos Aerial Vehicles

Time-constrained graph searches with provable bounds on suboptimality

Anytime graph searches for planning under time constraints. For example, ARA* (NIPS '03)

Incremental graph searches for planning in unknown and/or dynamic environments. For example, LPA* (NIPS '02), D* Lite (Transactions on Robotics and Automation '05), Real-Time Adaptive A* (AAMAS '06).

Anytime incremental graph searches for planning both under time constraints and in unknown and/or dynamic environment. For example, Anytime D* (Artificial Intelligence Journal '08).

Randomized graph searches for planning in high-dim. spaces with provable bounds on sub-optimality. For example, R* (AAAI '08)

*Application of Anytime D to planning dynamically-feasible paths for various robotic systems:**
Example of a planned path. Paper on planning long dynamically-feasible motions (RSS '08).	Winner of the DARPA Urban Challenge '07 (CMU vehicle). movie: motion planning with Anytime D* during the race (over 100MB)	ATRV navigating an unknown parking lot. movies: run, map building and re-planning	Segbot navigating unknown spaces. movie	Motion planning for unmanned helicopters. Ongoing project with DPI.

*Application of ARA and R* to path/motion planning in high-dimensional spaces:**
Planning with ARA* for 6 DOF robot arm. movie	Planning with ARA* for 20 DOF robot arm. movie	Planning sequence of steps with R* for a robotic quadruped. movie	Motion planning for the arms of a household robot. Ongoing project with Willow Garage. real robot movie

Large-scale decision-theoretic planning under uncertainty with graph searches

Probabilistic planning with Clear Preferences. For example, PPCP (Artificial Intelligence Journal '09)

Planning for Markov Decision Processes with Sparse Stochasticity. For example, MCP (NIPS '04)

Applications:
Robot-helicopter coordination with MCP. movie	Path Clearance with 5 UAV scouts using PPCP. movie	Autonomous landing with PPCP. Ongoing project with DPI.