RHCP Comparison Dataset Jan26

RHCP Comparison Dataset Jan 2026This dataset contains information relating to the 1800 motion plans generated and evaluated as part of the article "A receding-horizon multi-contact motion planner for legged robots in challenging environments".What is in this datasetAs discussed in the article, the RHCP and CVBFP motion planners were each used to plan several motions for the Corin hexapod in each of four simulated scenarios. The RHCP tests were repeated for horizon depths of 1, 2, 3 and 4 steps. CVBFP, and RHCP with horizon depths of 1, 2 and 3, each generated 100 motions per scenario. Due to the long planning times involved, only 50 motions were planned by RHCP with a 4 step horizon.For each run in a given environment, a partially randomised starting configuration was sampled. Every planner was then used to generate a motion plan from that sampled starting configuration to the same goal position. In each instance, the motion planning logs were stored and a number of metrics were measured for each plan. These metrics include the execution time, number of posture generation calls, and number of stance changes in the plan. When interpreting the stance change data, note that for CVBFP this records the number of stance changes, but for RHCP it records the number of steps (each step contains two stance changes, one for breaking the existing contact and one for forming a new one), so the RHCP data should be doubled for a fair comparison (this is already the case for the figures given in the article).Motion planning attempts whose planning time exceeded two hours were terminated without returning a plan. These entries have their planning time recorded as "Timeout" and are excluded from the analysis in the article. Since these runs did not return a motion plan, some metrics (<i>e.g.</i>, number of stance changes in the plan) are recorded as "N/A".Organisation of the dataThe top level of this directory contains the file <code>Experiment_Metric_Summary.csv</code>, which contains summary data about every motion plan generated (including the metrics above). This formed the basis of the statistical analysis presented in the paper.The rest of the data set is divided into four top level directories, with one for each scenario in which testing took place (<code>chimney_climbing</code>, <code>chimney_walking</code>, <code>stepping_stones</code>, <code>wall_walking</code>). Each directory contains a log of the randomly sampled starting configurations (<code>_starting_configs_Log.log</code>) as well as several sub-directories (one for each planner) containing the planning log files. The directory <code>/CVBFP </code>contains the CVBFP planning logs for that scenario, while the directory <code>/RHCP_X</code> contains the planning logs for RHCP with a X-step horizon depth.Each planning run generated a data log and an events log. The data log contains a summary of each posture generation call, recording information such as the generation time, validity, and potential of each node. The events log is a time stamped log of events during the planning process, recording the order in which nodes were selected for expansion, any calls to the intervention mechanism, any exceptions that occurred, <i>etc.</i>.NotesNote that the timestamps in each events log reflect when those log messages were transcribed into the log file, rather than when the events described occurred. Posture generator calls are typically made concurrently by several threads, and log messages are transcribed only after all threads have merged at the conclusion of a batch of calls. Thus, the time taken for nodes to be generated cannot be accurately judged from the event log timestamps. This information is recorded instead in the data logs.This dataset is discussed in more detail in the PhD thesis of Daniel S. J. Derwent (formerly Daniel Johnson).The unzipped file size of the dataset is approximately 48.6 GB. If only the statistical measures for each run are of interest, rather than the planning logs, then only the file <code>Experiment_Metric_Summary.csv</code> need be extracted.