Dave.Touretzky@C.CS.CMU.EDU (05/06/88)
For the record: Dean's preliminary experiments on *simulated* road images give very good results. His experiments on 16 *real* road images also gave good results, but that's way too few images to say anything conclusive about how his net will compare with the CMU vision group's software. While I personally think that the neural net approach can eventually outperform other approaches on this simple road following task, and will do so shortly, we don't have the data in hand to make public claims like that. Also, the ALV task involves more than just road tracking. One must deal with things like forks and intersections in roads; one must be able to use a map to keep track of current position and navigate from starting points to arbitrary destinations. Neural nets have not yet even been applied to these more complex tasks. It's anybody's guess how well they'll perform. -- Dave Touretzky -------