Non-Monotonic Reasoning on Board a Sony AIBO

摘要

One of the most satisfying aspects of our implementation is that it has proven efficient enough to be running on board a Sony AIBO while in competition in the soccer 4-legged league or in RoboCup@Home. Table 1 shows the CPU-timings on board an ERS-7 running our C++ implementation with three different models and two situations. It may be surprising that while the robot was in the playing state, chasing the ball and executing kicks, the execution is faster than while standing as a goalie. However, the standing situations have usually on average 2 landmarks per frame. But while playing, frames with 2 objects in sight are less frequent. In all cases, the inference engine is executed six times per frame, to verify if each of the landmarks is consistent. Model Activity Phase 1 and 95 % Phase 1, 95 % Net Phase 3 Phase 2 Confidence Phase 2, and Confidence Interval Phase 3 Interval 749 s ? 7 s 931 s ? 8 s 182 s 1,438 s ? 31 s 1,687 s ? 35 s 249 s 407 s ? 15 s 622 s ? 17 s 215 s 209 s ? 13 s 371 s ? 17 s 162 s