Abstract | ||
---|---|---|
We present a neural network aided exploration strategy with the goal of improving the performance of robot-only mapping and exploration by considering geometric and topological features around the exploration frontier. Using an information-theoretic exploration strategy, we introduce learning in the selection of next best candidate frontier locations for exploration. Geometric and topological knowledge of the space surrounding candidate locations provided by a domain expert is used to train a fully-connected neural network offline. The result is then applied by the robot in its exploration of new environments. Our results show that in addition to metric information about the environment, the robot can learn to apply domain knowledge in its selection of next best frontier locations. The result is an increase in the rate of acquisition of topological information for a given space. |
Year | DOI | Venue |
---|---|---|
2018 | 10.1109/SSRR.2018.8468650 | 2018 IEEE International Symposium on Safety, Security, and Rescue Robotics (SSRR) |
Keywords | Field | DocType |
topological features,exploration frontier,information-theoretic exploration strategy,candidate frontier locations,topological knowledge,metric information,topological information,robot-only mapping,fully-connected neural network,robot-only exploration,learning,neural network offline training | Computer vision,Topological information,Domain knowledge,Computer science,Subject-matter expert,Artificial intelligence,Robot,Artificial neural network | Conference |
ISBN | Citations | PageRank |
978-1-5386-5573-3 | 0 | 0.34 |
References | Authors | |
6 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Hantian Liu | 1 | 0 | 0.34 |
M. Ani Hsieh | 2 | 382 | 34.69 |