Robust Neuro-Symbolic Goal and Plan Recognition

Goal Recognition is the task of discerning intended goal an agent given a sequence observations, whereas Plan consists identifying plan to achieve such goal. Regardless underlying techniques, most recognition approaches are directly affected by quality available observations. In this paper, we develop neuro-symbolic that can combine learning and planning compensating for noise missing observations using prior data. We evaluate our in standard human-designed domains as well domain models automatically learned from real-world Empirical experimentation shows reliably infer goals compute correct plans experimental datasets. An ablation study outperform rely exclusively on model, or machine problems with both noisy low observability.