POMDP Planning for Object Search in Partially Unknown Environment

Authors: Yongbo Chen, Hanna Kurniawati

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We tested our approach using Gazebo simulations on four scenarios of target finding in a realistic indoor living environment with the Fetch robot simulator. Compared to the baseline approaches, which are based on POMCP, our results indicate that our approach enables the robot to find the target object with a higher success rate faster while using the same computational requirements.
Researcher Affiliation Academia Yongbo Chen School of Computing, Australian National University Canberra, ACT, 2601, Australia Yongbo.Chen@anu.edu.au Hanna Kurniawati School of Computing, Australian National University Canberra, ACT, 2601, Australia Hanna.Kurniawati@anu.edu.au
Pseudocode Yes The pseudocode Algorithms of the GPOMCP solver are shown in supplementary materials.
Open Source Code No The paper states that pseudocode is in the supplementary materials, but there is no explicit statement about releasing the source code or a link to a code repository for the described methodology.
Open Datasets No The paper describes setting up scenarios in a Gazebo simulator and using self-generated maps and object configurations. It does not provide concrete access information (specific link, DOI, repository name, or formal citation with authors/year) for a publicly available or open dataset used for training or evaluation.
Dataset Splits No The paper describes conducting 20 trials for each scenario in a simulation environment and reporting statistical results, but it does not specify explicit dataset splits (e.g., percentages, sample counts, or citations to predefined splits) for training, validation, or testing data.
Hardware Specification Yes The project operates on multiple Ubuntu 18.04 desktops equipped with an Intel Core i7-13700k processor, running exclusively on CPU.
Software Dependencies No The paper mentions several software components like 'C++ and Python codes', 'ROS interface move_base', 'ros_control [5]', 'ROS tool moveit [6]', 'YOLO detection', 'SIFT matching [17]', 'GPD toolbox [23]', 'k-means clustering algorithm', 'RTAB-Map [16]', and 'AMCL-based navigation stack [18]', but it does not provide specific version numbers for any of them.
Experiment Setup Yes The planning time is limited to 60 s/step, the discounted factor γ = 0.9, the threshold to the object declaration Ct d = log(8/2) and Co d = log(2/8), the thresholds for grid updating νp = 0.1 and νn = 0.1, the re-initialized grid values of the guessed target object are set as 0.2, the reward values Rmax = 105, Rct = 5 104, Rco = 104, Rmin = 1, and Rill = 103. The grid size of the grid world is set as 2 cm. A discussion about different grid sizes is shown in the supplementary materials. The log-odds value of these grids is initially set as unknown satisfying log 0.5 1 0.5 = 0.