DeepFoids: Adaptive Bio-Inspired Fish Simulation with Deep Reinforcement Learning
Authors: Yuko Ishiwaka, Xiao Zeng, Shun Ogawa, Donovan Westwater, Tadayuki Tone, Masaki Nakada
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In Sec. 5, we show the experimental results and analysis followed by a conclusion in Sec. 6. We trained the behavioral models of coho salmon, yellowtail and red seabream. We compare the results of simulations trained by PPO and SAC. The results of the experiments with each component of the environment simulation are illustrated in Fig. 10. |
| Researcher Affiliation | Industry | Yuko Ishiwaka1 , Xiao S. Zeng2 , Shun Ogawa1, Donovan Michael Westwater2, Tadayuki Tone1, Masaki Nakada2 1Soft Bank Corp., Japan, 2 Neural X Inc., USA |
| Pseudocode | No | The paper describes the algorithms and processes in text and with diagrams (e.g., Figure 1 for simulation framework, Figure 2 for policy network architecture), but it does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | The paper mentions utilizing third-party tools like Unity Engine with ML-Agents toolkit and a modified YOLOv4 model, but it does not provide a specific link or explicit statement about the availability of the authors' own source code for the described methodology. |
| Open Datasets | No | The paper explicitly states that it created a 'high-quality synthetic dataset' using its simulation (Fig. 3 displays an example), but it does not provide any access information (link, DOI, repository, or citation) for this dataset to be publicly available. |
| Dataset Splits | No | The paper describes a two-step training scheme for the behavioral models and discusses pretraining and transfer learning phases. However, it does not provide explicit details about training/validation/test dataset splits, specific percentages, or sample counts needed to reproduce the data partitioning. |
| Hardware Specification | Yes | The training took place on an NVIDIA GeForce RTX 3080 Laptop GPU installed on a Windows 10 machine with a 3.3 GHz AMD Ryzen 9 5900HX CPU. |
| Software Dependencies | No | The paper mentions key software components such as 'Unity Engine with ML-Agents toolkit', 'Py Torch', and 'YOLOv4', but it does not specify version numbers for these software dependencies. |
| Experiment Setup | Yes | For yellowtail and red seabream, we ran a total number of two million time steps and used a linearly decaying learning rate of 0.001 in both phases. For coho salmon, we adopted the same settings, but increased the pretraining and transfer learning steps to four million and one million time steps respectively. We enabled early termination of an episode if the agent collides with either a cage wall or the water surface. |