Technical Appendix 1 Anubhav Singh, 1 Nir Lipovetzky, 2 Miquel Ramirez, 3 Javier Segovia-Aguas 1 School of Computing and Information Systems, University of Melbourne, Australia 2 Electrical and Electronic Engineering, University of Melbourne, Australia 3 Dept. Information and Communication Technologies, Universitat Pompeu Fabra, Spain anubhavs@student.unimelb.edu.au, {nir.lipovetzky, miquel.ramirez}@unimelb.edu.au, javier.segovia@upf.edu 1 Implementation details Hash Functions We used the std::hash method from the standard C++ library to compute the hash values. The hash value were combined using the method hash combine de- scribed in the proposal for C++ standard N3876 (Josuttis 2014). Seeds used for multiple runs of planners using random- ization We used the sequence 101, 204, 307 and 410, formed by concatenating two AP sequences of the form x 1 =1+(n - 1) and x 2 = 01 + 3(n - 1) , for n ∈ [1, 4]. 2 Additional results 2.1 P 2 (BFWS(f 5 )) vs. pI -P ¯ ω AC : Coverage with the Agile track constraints on time and memory From Table. 1, we observe that the coverage of pI -P ¯ ω AC is 59 higher than P 2 (BFWS(f 5 )), one of the best performing to-date in the Agile track (IPC 2018). The experiments were performed with the time and memory limit of 300 sec and 8 GB respectively. BFWS(f 5 ) pI -P ¯ ω AC BFWS(f 5 ) 0 78 pI -P ¯ ω AC 19 0 Table 1: Comparing BFWS(f5) and pI -P ¯ ωAC. The value in a cell represents the number of instances which were solved by the col planner but not the row planner, with the time limit of 300 sec and memory limit of 8 GB. 2.2 Pairwise comparison of plan cost between planners using Approximate Novelty Search and P 2 (BFWS(f 5 )) From Fig. 1 and 2, we note that the plan length remain sim- ilar for different configuration of BFWS( ˆ f 5 ) planners using approximate novelty search. 10 1 10 2 10 3 10 4 BFWS(f5) 10 1 10 2 10 3 10 4 P 2 AC Plan cost Figure 1: Pairwise comparison of cost between BFWS(f5) and P2AC on instances from every IPC satisficing benchmark. 10 1 10 2 10 3 10 4 BFWS(f5) 10 1 10 2 10 3 10 4 pI -P ¯ ω AC Plan cost Figure 2: Pairwise comparison of cost between BFWS(f5) and pI - P ¯ ωAC on instances from every IPC satisficing benchmark. 2.3 Complete table of results In Table. 2, we show the coverage on all the domains which were part of our experiments. We included all the domains Copyright © 2021, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.