Learning Space Partitions for Path Planning
Authors: Kevin Yang, Tianjun Zhang, Chris Cummins, Brandon Cui, Benoit Steiner, Linnan Wang, Joseph E. Gonzalez, Dan Klein, Yuandong Tian
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, La P3 outperforms existing path planning methods in 2D navigation tasks, especially in the presence of difficult-to-escape local optima, and shows benefits when plugged into the planning components of model-based RL such as PETS [7]. These gains transfer to highly multimodal real-world tasks, where we outperform strong baselines in compiler phase ordering by up to 39% on average across 9 tasks, and in molecular design by up to 0.4 on properties on a 0-1 scale. |
| Researcher Affiliation | Collaboration | 1UC Berkeley 2Facebook AI Research 3Brown University |
| Pseudocode | Yes | Algorithm 1 La P3 Pseudocode for Path Planning. Improvements over La MCTS in green. |
| Open Source Code | Yes | Code is available at https://github.com/yangkevin2/neurips2021-lap3. |
| Open Datasets | Yes | We use Mini World [5] for continuous path planning and Mini Grid [6] for discrete. ... For DRD2, HIV, and SARS, we evaluate using computational predictors from [33] (DRD2) and [51] (HIV, SARS) in lieu of wet-lab assays. ... learn a latent representation from a subset of Ch EMBL [29], a dataset of 1.8 million drug-like molecules. |
| Dataset Splits | No | The paper does not explicitly specify exact percentages or counts for training, validation, and test splits. It mentions running multiple trials and using a sliding window for success percentage, but not a general dataset splitting methodology. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used for running the experiments. It states in the ethics checklist: "(d) Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [No]" |
| Software Dependencies | No | The paper does not provide specific software names along with their version numbers (e.g., Python 3.8, PyTorch 1.9) required for replication. |
| Experiment Setup | Yes | Parameters: Initial #samples Ninit, Re-partitioning interval Npar, Node partition threshold Nthres, UCB parameter Cp. ... The only additional hyperparameter tuned in La P3 is the Cp controlling exploration when selecting regions to sample from, which is dependent on the scale of the reward function. However, our Cp only varies by a factor of up to 10 across our diverse environments, and performance is not overly sensitive to small changes (Appendix F.5). |