Learning to Infer Final Plans in Human Team Planning

Authors: Joseph Kim, Matthew E. Woicik, Matthew C. Gombolay, Sung-Hyun Son, Julie A. Shah

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our technique on human team planning datasets across four different planning domains. We demonstrate that our technique infers teams final plans with higher accuracy than prior art. With our domain-independent feature set, we empirically demonstrate that our learned model can be applied to successfully infer team plans within a novel plan- ning domain, and that performance is robust to incomplete planning domain models.
Researcher Affiliation Collaboration 1 Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology 2 MIT Lincoln Laboratory , Lexington, MA
Pseudocode Yes Algorithm 1 Policy gradient algorithm for plan inference.
Open Source Code No The paper does not provide any statement about releasing its source code or a link to a code repository for the methodology described.
Open Datasets Yes We tested our model on a human team planning dataset collected by Kim and Shah [2016], wherein teams of two participants each conversed via web chat and generated plans for a hypothetical emergency response scenario.
Dataset Splits Yes λ was set by maximizing F1-score using 10% of the documents as a validation set.
Hardware Specification Yes All tests were performed using an Intel(R) Xeon(R) E5-2630 v4 processor (2.20 GHz, 24 cores) with 48 GB RAM.
Software Dependencies No The paper mentions using "LAMA [Richter and Westphal, 2010]" and "Metric-FF [Hoffmann, 2003]" as planners, but does not provide specific version numbers for these or any other ancillary software components.
Experiment Setup Yes Each of these trials used T=500 and α of 0.005. Our policy gradient algorithm uses the standard ϵ-greedy exploration, with ϵ decay rate of 1/(10 + t).