Loading...
Please wait, while we are loading the content...
Similar Documents
Efficient Learning of Trajectory Preferences Using Combined Ratings and Rankings
| Content Provider | Semantic Scholar |
|---|---|
| Author | Somers, Thane Lawrance, Nicholas R. J. Hollinger, Geoffrey A. |
| Copyright Year | 2017 |
| Abstract | In this paper we propose an approach for modeling and learning human preferences using a combination of absolute (querying an expert for a numerical value) and relative (asking the expert to select the highest-value option from a set) queries. Our approach uses a Gaussian process regression model with an associated likelihood function that can take into account both pairwise preferences and numerical ratings to approximate the user’s latent value function from a set of (noisy) queries. We show that using a combination of relative and absolute queries performs better than either query type alone and propose a simple active learning approach to sequentially select informative queries that speed up the learning process when searching for high-value regions of the user’s latent value space. We demonstrate the effectiveness of our method on a 1-D function approximation task and on a simulated autonomous surface vehicle performing a lake monitoring mission. These experiments show that our algorithm is able to efficiently learn an operator’s mission preferences and use those mission preferences to autonomously plan trajectories that fulfill the operator’s goals. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://research.engr.oregonstate.edu/rdml/sites/research.engr.oregonstate.edu.rdml/files/somersrss17.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
