Fig. 5: Anomaly active target tracking performance of all agents in different test scenarios.
From: Cognitive embodied learning for anomaly active target tracking
a, b display the success rate (SR) and the relative path length (RPL) for different agents in the Test Scenario-1. c–f display variations of SR and RPL across test scenarios-2 and test scenarios-3 among agents using rules versus those without, and among agents employing different rules. EL(EA + SI), EL(EA), EL(SI), and EL(STF) represent four common embodied learning (EL) paradigms. EA + SI is a policy learning method that uses the semantic feature cognition (SFC) and the sequential cognitive memory (SCM) to process observations. EA is a policy learning method that uses the SFC to process observations. SI is a transformer-based method that uses the SFC and the SCM to generate semantic tokens. STF is an adaptive embodied learning method based on the student-teacher framework. NBV is a region-based viewpoint optimization computational method. Error bars represent standard deviation (SD) across ten independent runs. Each run corresponds to an evaluation with a different random seed.
