|
|
Foundations of Robotics
Seminar, May 6, 2009
Time
and Place | Seminar Abstract
ICRA 2009 Practice Talks
Onboard Contextual Classification of 3-D Point Clouds with Learned
High-order Markov Random Fields
Daniel Munoz
Carnegie Mellon University - Robotics Institute
Active Guidance of a Handheld Micromanipulator using Visual Servoing
Brian Becker
Carnegie Mellon University - Robotics Institute
Path Diversity Is Only Part of the Problem
Ross Alan Knepper
Carnegie Mellon University - Robotics Institute
NSH 1507
Talk 4:30 pm
Onboard Contextual Classification of 3-D Point Clouds with Learned
High-order Markov Random Fields
Contextual reasoning through graphical models such as Markov Random Fields often show superior performance against local classifiers in many domains. Unfortunately, this performance increase is often at the cost of time consuming, memory intensive learning and slow inference at testing time. Structured prediction for 3-D point cloud classification is one example of such an application. In this paper we present two contributions. First we show how efficient learning of a random field with higher-order cliques can be achieved using subgradient optimization. Second, we present a context approximation using random fields with high-order cliques designed to make this model usable online, onboard a mobile vehicle for environment modeling. We obtained results with the mobile vehicle on a variety of terrains, at 1/3 Hz for a map 25 x 50 meters and a vehicle speed of 1-2 m/s.
Active Guidance of a Handheld Micromanipulator using Visual Servoing
In microsurgery, a surgeon often deals with anatomical structures of sizes that are close to the limit of the human hand accuracy. This talk demonstrates control of a handheld tremor reduction micromanipulator with simple visual servo techniques, aiding the operator by providing three behaviors: snap-to, motion-scaling, and standoff-regulation. A stereo camera setup viewing the workspace under high magnification tracks the tip of the micromanipulator and the desired target object. We show that the snap-to behavior can reach and maintain a position at a target with an accuracy several times greater than an unaided human.
Path Diversity Is Only Part of the Problem
Path sets, a technique for local motion planning, fit into a hierarchical planning framework in which a group of planners operate at different range and fidelity to quickly produce real-world motion plans. Path sets sample a small number of possible short-range trajectories using a high-fidelity simulator. A path-set-based planner must replan frequently to compensate for its coarse sampling of path space. Path sets are frequently compared using path diversity, which is akin to dispersion. In this talk, I explore two different techniques for managing path sets. In the first, a path set is held fixed to the moving body of the vehicle, which only ever executes a small fraction of any path. The second technique involves fixing a path set to the ground instead, and allowing the robot to traverse entire paths. In addition to exploring the advantages and disadvantages of each approach, I will demonstrate through empirical evidence that while the performance of any path set is based on its constituent paths, performance in either approach is a poor predictor of performance in the alternative approach. This implies that there are more factors influencing planning performance than just path diversity.
The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.