We have rewritten some of the function primitives in the Next-Best-View program to deal better with floating-point and noise issues that appear under normal robot operations. Numerical uncertainties always cause errors in any system, but the new primitives prevent the errors from spreading or becoming catastrophic.
We implemented a number of heuristics in order to prune alignment candidates before performing a full comparison between two scans. This results in a speedup of 5 to 10 times in the alignment process, and is more robust as well.
We successfully mapped a section of our lab cluttered with many obstacles (chairs, tables, other robots).
We have equipped another of our robot as a map builder, with a fast serial PCI card that allows measurement speeds of 32 scans per second. We have constructed a simple aluminum mount that will allow the robot to take 3D images in three projection modes: spherical, pseudo-orthogonal lateral, and pseudo-orthogonal upward. Our goal is now to have one robot perform 2D map building while the second one performs 3D scans. We are also planning to evaluate collaborative map building strategies.
We have been investigating the possibility of extracting high-level symbolic objects for the interior of a building from 3D scans, using Bayesian estimation methods.The goal is to reduce the amount of data needed to represent a 3D environment. By symbolic objects we mean, for example, walls, doors, ceiling, etc.
We have finished implementing a version of the two observers/two targets planner, and ran it in simulation. What is interesting in our approach is that there is no predetermined assignment of a given target to a given observer. At any instant in time, the two observers place themselves so as to maximize the time it would take for any of the targets to escape.
We are working on a number of cases where there is an undesirable `shadowing' effect: if one observer sees both targets and the other observer is far away, the latter just remains where it is instead of coming to help.
The software architecture, that is, the communication protocols between the observers and the central planner, is already in place. We have to equip a new robot with the appropriate sensing devices, and adapt the vision part of the system so that an observer can detect two targets at once.
Nomadic Technologies sent us their new DC-wiring kits which correct a bug in the original Super Scout design. The bug caused the motor amplifiers to burn when the Scout was commanded to stop after moving at the specified maximum distance. Although we never lost an amplifier due to the bug, the upgrade will enhance the performance of the target tracker.
We produced a preliminary report with quantitative comparisons between a human operator and a computer on certain essential components of our programs.