Quarterly Progress Report on Contract N00014-91-J-1577 (Yale University)
Abstract
During this period, we continued our work on perception and planning for autonomous agents. (1) We designed and implemented algorithms for visual tracking and for grasping objects of unknown shape from visual data. These are being written up. (2) We proved the correctness and completeness of a decision- making algorithm based on interval analysis. We were able to show that the procedure is essentially a decision procedure except for a very small set of cases that almost never arise in practice. These results were written up and submitted to Interval Computations. (3) We redesigned and reimplemented our time map for projecting robot plans. The biggest enhancement was to add Poisson- distributed autonomous events. You can now model the actions of other agents with rules like (p - > E A t B), which states that over any time interval where A is true, the average time to the next occurrence of B is t. The next problem is to prove the algorithm correct; technical subtleties arise. (4) We have begun to rethink our notion of runtime planning. Currently the planner executes its entire plan simultaneously with the attempt to improve it. If the improvement attempt succeeds, the new version is swapped in. What we would like to do is provide alternative ways to interleave planning and execution, so that some planning can be deferred until runtime. We are exploring how to add a RUNTIME- PLAN statement to RPL, our plan language.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1993
- Accession Number
- ADA268447
Entities
People
- Drew McDermott
Organizations
- Yale University