Information Maximizing Curriculum: A Curriculum-Based Approach for Learning Versatile Skills
Authors: Denis Blessing, Onur Celik, Xiaogang Jia, Moritz Reuss, Maximilian Li, Rudolf Lioutikov, Gerhard Neumann
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness of our approach on complex simulated control tasks using diverse human demonstrations, achieving superior performance compared to state-of-the-art methods. (Abstract) and 5 Experiments (Section title) |
| Researcher Affiliation | Academia | Denis Blessing i Onur Celik Xiaogang Jia Moritz Reuss Maximilian Xiling Li Rudolf Lioutikov Gerhard Neumann Autonomous Learning Robots, Karlsruhe Institute of Technology Intuitive Robots Lab, Karlsruhe Institute of Technology FZI Research Center for Information Technology Correspondence to denis.blessing@kit.edu |
| Pseudocode | Yes | Algorithm 1 IMC training procedure (Section 4, page 4) |
| Open Source Code | Yes | The code is available online 2https://github.com/ALRhub/imc (Section 5, page 5) |
| Open Datasets | No | The paper describes the datasets used (e.g., "The obstacle avoidance dataset contains four human demonstrations... amounting to 7.3k (o, a) pairs." in Section 5.1), but it does not provide specific URLs, DOIs, repository names, or formal citations with authors/year for public access to these collected datasets. |
| Dataset Splits | No | The paper discusses training and evaluation but does not specify exact dataset split percentages (e.g., "80/10/10 split") or absolute sample counts for training, validation, and testing, nor does it cite predefined standard splits that include validation. |
| Hardware Specification | No | The paper mentions the use of "bw HPC, as well as the Hore Ka supercomputer" in the Acknowledgments section, which indicates computational resources were used. However, it does not provide specific hardware details such as GPU models, CPU models, or memory configurations used for running the experiments. |
| Software Dependencies | No | The paper mentions general software components like "Mu Jo Co physics engine" (Section 5.1) and specific libraries by name (e.g., "PyTorch" implied by neural networks and "Bayesian optimization" [49]), but it does not provide specific version numbers for these software dependencies (e.g., "Python 3.8, PyTorch 1.9"). |
| Experiment Setup | Yes | For all experiments, we use batch-size |B| = |D|, number of components Nz = 50 and expert learning rate equal to 5 10 4. Furthermore, we initialized all curriculum weights as p(on, an|z) = 1. For the table tennis and obstacle avoidance task, we found the best results using a multi-head expert parameterization (see Section E.2) where we tested 1 4 layer neural networks. We found that using 1 layer with 32 neurons performs best on the table tennis task and 2 layer with 64 neurons for the obstacle avoidance task. (Section C.2) and Tables 2, 3, 4, 5, 6, 7, 8. |