I'm not speaking here of noisy robots. SLAM is an acronym for 'simultaneous localization and mapping,' and slamming robots are simply robots which at the same time build a map of their environment while keeping track of their own localization. Now, according to New Scientist, Purdue University engineers have designed P-SLAM robots based on a prediction-based SLAM algorithm. This algorithm contains an environmental-structure predictor to predict the structure inside an unexplored region. As these robots make some guesses about their environments, they are able to react faster in unfamiliar buildings. But if this technique is working well in offices -- which have some repetitive characteristics from room to room -- it is not yet well-adapted to outdoor environments.
 These algorithms have been developed at Purdue University's Distributed Energy-efficient Autonomous Robots (DEAR) lab, led by professor George Lee and assistant professor Y. Charlie Hu.
On the left, you can see several final maps generated by an Rao–Blackwell particle filter (RBPF) on the top and by P-SLAM in the middle. "The inconsistent global map [obtained with RBBF] was caused by incorrect angle estimations of the corners. The reason is that when the robot was turning, it injected some large rotation uncertainty into the system state. If the number of particles is not large enough to cover the spread of the system state, the particle filter will not be able to close the loop. On the other hand, [in the middle image,] P-SLAM successfully closed the loop due to correctly predicting the corner structures, which means that P-SLAM reduced the required number of particles. To see how many particles are needed to obtain a convergence result in an RBPF, we increased the number of particles by 100 each time. The offline simulation results showed that we needed 1000 particles to obtain the converged mapping result, which is shown in [the bottom image]." (Credit: Purdue University)
Now, here are some quotes from the New Scientist article.
The team's algorithm identifies unexplored regions, known as "frontier cells", adjacent to areas that have already been mapped. It then uses the pattern of corners at the edge of this cell to search for similar patterns that have already been mapped. If a match is found, the algorithm uses the existing map to make a prediction of the contents of the frontier cell. Each prediction has a "confidence score" attached to it. Areas with a high score can be left unexplored to save time, while predictions with lower confidence scores may need to be mapped properly.
The algorithm was initially tested using simulated robots, placed inside virtual mazes and office environments. The simulated robots were able to navigate successfully while exploring 33% less of their environment. Real-life tests where then carried out with small robots inside an office building at the university. The real robots also saved time and experienced fewer mapping errors, thanks to combined predictions with measurements made using onboard instruments. Building maps using measurements alone is less accurate because instruments are prone to errors.
This research work has been published by IEEE transactions on robotics under the name "P-SLAM: Simultaneous Localization and Mapping With Environmental-Structure Prediction" (Volume 23, Issue 2, Pages 281-293, April 2007). Here is a link to the abstract which provides additional details.
The prediction process is based on the observation of the surroundings of an unexplored region and comparing it with the built map of explored regions. If a similar environment/structure is matched in the map of explored regions, a hypothesis is generated to indicate that a similar structure has been explored before. If the environment has repeated structures, the mobile robot can use the predicted structure as a virtual mapping, and decide whether or not to explore the unexplored region to save the exploration time. If the mobile robot decides to explore the unexplored region, a correct prediction can be used to speed up the SLAM process and build a more accurate map.
For even more information, you should read the full technical paper (PDF format, 13 pages, 2.06 MB), from which the above images and caption have been extracted.
Sources: Tom Simonite, New Scientist, May 8, 2007; and various websites
You'll find related stories by following the links below.
7:08:48 PM
Permalink
|
|
