Optimal and Efficient Stochastic Motion Planning in Partially-Known Environments

Authors: Ryan Luna, Morteza Lahijanian, Mark Moll, Lydia Kavraki

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments confirm that the framework recomputes high-quality policies in seconds and is orders of magnitude faster than existing methods. Experiments are conducted in a continuous 20x20 maze environment (Figure 2) to evaluate the computation time and quality of the resulting policy (probability of success) for the three BMDP update methods when new obstacles are observed.
Researcher Affiliation Academia Ryan Luna, Morteza Lahijanian, Mark Moll, and Lydia E. Kavraki Department of Computer Science, Rice University Houston, Texas 77005, USA {rluna, morteza, mmoll, kavraki}@rice.edu
Pseudocode No The paper describes algorithmic concepts but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statement about releasing source code for the methodology or a link to a code repository.
Open Datasets No The paper describes experiments in a 'continuous 20x20 maze environment (Figure 2)' which appears to be a custom setup, and no access information (link, DOI, specific repository, or citation to a public dataset) is provided.
Dataset Splits No The paper describes a simulated environment and experiment setup but does not specify dataset splits (e.g., percentages, sample counts) for training, validation, or testing.
Hardware Specification Yes All computations are performed using a 2.4 GHz quad-core Intel Xeon (Nahalem) CPU.
Software Dependencies No The paper mentions algorithms and methods (e.g., iMDP, IVI) but does not provide specific software dependencies with version numbers (e.g., 'Python 3.8', 'PyTorch 1.9').
Experiment Setup Yes For BMDP construction, the maze is discretized using a Delaunay triangulation that respects known obstacles, where no triangle exceeds more than 0.1% of the free space, resulting in 759 discrete regions. Three local policies are computed within each triangle until there are 1000 states per unit area in each policy. The system evaluated has 2D single integrator dynamics with Gaussian noise. The dynamics are in the form of (1), where f(x, u) = u and F(x, u) = 0.1I, and I is the identity matrix. The system receives a terminal reward of 1 when it reaches the goal and a terminal reward of 0 for colliding with an obstacle. The reward rate for all non-terminal states is zero (g = 0), and a discount rate of 0.95 is employed for the local policies. The system can perfectly sense all unknown features within a radius rDETECT of 2 unit length.