Fourier Controller Networks for Real-Time Decision-Making in Embodied Learning
Authors: Hengkai Tan, Songming Liu, Kai Ma, Chengyang Ying, Xingxing Zhang, Hang Su, Jun Zhu
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive results in both simulated (e.g., D4RL) and real-world environments (e.g., robot locomotion) demonstrate FCNet s substantial efficiency and effectiveness over existing methods such as Transformer, e.g., FCNet outperforms Transformer on multienvironmental robotics datasets of all types of sizes (from 1.9M to 120M). In this section, we conduct extensive experiments to evaluate the performance and efficiency of FCNet. |
| Researcher Affiliation | Collaboration | 1Department of Computer Science and Technology, Institute for AI, Tsinghua-Bosch Joint ML Center, THBI Lab, BNRist Center, Tsinghua University, Beijing, 100084, China. |
| Pseudocode | Yes | A detailed schematic of the CSC block, illustrating the computation of y0, y1, . . . , yn 1, is shown in Fig. 7. |
| Open Source Code | Yes | The project page and code can be found https://thkkk.github.io/fcnet. |
| Open Datasets | Yes | First, in the classic offline RL environments such as D4RL (Fu et al., 2020)... The multi-environment legged robot locomotion dataset is an expert dataset, collected in Isaacgym (Makoviychuk et al., 2021)... We make some of the data public on the project page. |
| Dataset Splits | Yes | To evaluate FCNet in the D4RL (Fu et al., 2020) offline RL tasks... We report the performances of FCNet and Transformer across various dataset sizes in Fig. 4... The test version of the dataset contains the aforementioned skills and terrains, with 320,000 trajectories and totaling 60M steps. |
| Hardware Specification | Yes | Our test environment is shown in Table 8: CPU Intel(R) Xeon(R) Silver 4210 CPU @ 2.20GHz |
| Software Dependencies | Yes | Our test environment is shown in Table 8: Python 3.8.18 GCC 8.4.0 Torch Version 1.12.1 |
| Experiment Setup | Yes | The hyperparameters of FCNet for D4RL are shown in Table 4. The hyperparameters of FCNet for the multi-environment legged robot locomotion dataset are shown in Table 5. The hyperparameters of Transformer for the multi-environment legged robot locomotion dataset are shown in Table 6. In the experiments of Fig. 6, when we change a hyperparameter, other hyperparameters remain at their default values: Context Length= 64, n_layer= 4, hidden_size= 256. |