Accelerating Motion Planning via Optimal Transport
Authors: An T. Le, Georgia Chalvatzaki, Armin Biess, Jan R. Peters
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Further, our empirical evaluations on representative tasks with high-dimensionality and multimodal planning objectives demonstrate an increased benefit of MPOT, both in terms of planning time and success rate, compared to notable trajectory optimization methods. We experimentally evaluate MPOT in Py Bullet simulated tasks, which involve high-dimensional state space, multiple objectives, and challenging costs. |
| Researcher Affiliation | Academia | An T. Le1, Georgia Chalvatzaki1,3,5, Armin Biess, Jan Peters1 4 1Department of Computer Science, Technische Universitat Darmstadt, Germany 2German Research Center for AI (DFKI) 3Hessian.AI 4Centre for Cognitive Science 5Center for Mind, Brain and Behavior, Uni. Marburg and JLU Giessen, Germany |
| Pseudocode | Yes | Algorithm 1: Motion Planning via Optimal Transport |
| Open Source Code | No | The paper provides a link to a specific function in a GitHub repository ('https://github.com/anindex/ssax/ blob/main/ssax/ss/rotation.py#L38') for the Steward method in Appendix G, but not a general repository for the full MPOT methodology. A demo site is also linked, but not explicitly for the code. |
| Open Datasets | No | The paper describes custom-generated environments and scenarios for its experiments (e.g., 'we populate 15 square and circle obstacles randomly', 'randomly sampled 15 obstacle-spheres') rather than using named, publicly available datasets with concrete access information. |
| Dataset Splits | No | The paper describes the generation of experiment scenarios and tasks (e.g., '100 environment-seeds', '1000 planning tasks') but does not specify train/validation/test dataset splits or any methodology for reproducing such data partitioning for its own experiments. |
| Hardware Specification | Yes | All experiments are executed in a single RTX3080Ti GPU and a single AMD Ryzen 5900X CPU. |
| Software Dependencies | No | The paper mentions implementing baselines in 'Py Torch' but does not specify version numbers for PyTorch or any other software libraries or dependencies. |
| Experiment Setup | Yes | The MPOT hyperparameters used in the experiments are presented in Table 4. Table 4: Experiment hyperparameters of MPOT. α0, β0 are the initial stepsize and probe radius. h is the number of probe points per search direction. eps is the annealing rate. P is the polytope type, and λ is the entropic scaling of OT problem. |