Integrating Symmetry into Differentiable Planning with Steerable Convolutions

Authors: Linfeng Zhao, Xupeng Zhu, Lingzhi Kong, Robin Walters, Lawson L.S. Wong

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our approach on four tasks: 2D navigation, visual navigation, 2 degrees of freedom (2-DOF) configuration space manipulation, and 2DOF workspace manipulation. Our experimental results show that our symmetric planning algorithms significantly improve training efficiency and generalization performance compared to non-equivariant baselines, including VINs and GPPN.
Researcher Affiliation Academia Khoury College of Computer Sciences, Northeastern University
Pseudocode Yes D.2 PYTORCH-STYLE PSEUDOCODE
Open Source Code No We also plan to open source the codebase.
Open Datasets Yes For all environments, the planning domain is the 2D regular grid as in VIN, GPPN and SPT S = Ω= Z2, and the action space is to move in 4 directions1: A = (north, west, south, east). We randomly generate training, validation and test data of 10K/2K/2K maps for all map sizes, to demonstrate data efficiency and generalization ability of symmetric planning. For visual navigation, we randomly generate maps using the same strategy as before, and then render four egocentric panoramic views for each location from 3D environments produced with Gym-Mini World (Chevalier-Boisvert, 2018), which can generate 3D mazes with any layout.
Dataset Splits Yes We randomly generate training, validation and test data of 10K/2K/2K maps for all map sizes, to demonstrate data efficiency and generalization ability of symmetric planning.
Hardware Specification No The paper mentions general terms like "on a GPU" but does not specify any particular GPU model (e.g., NVIDIA A100), CPU model, or other detailed hardware specifications used for the experiments.
Software Dependencies No The paper mentions software components like "PyTorch-style pseudocode," the "e2cnn package," "U-net," and "Gym-Mini World," but it does not specify version numbers for any of these components.
Experiment Setup Yes The training process (on given maps) follows (Tamar et al., 2016a; Lee et al., 2018), where we train 30 epochs with batch size 32, and use kernel size F = 3 by default. The default batch size is 32. GPPN variants need smaller number because LSTM consumes much more memory. For non-Sym Plan related parameters, we use learning rate of 10 3, batch size of 32 if possible (GPPN variants need smaller), RMSprop optimizer. Due to memory constraints, we use K = 30 iterations for 2D maze navigation, and K = 27 for manipulation. We use kernel sizes F = {3, 5, 5} for m = {15, 28, 50} navigation, and F = {3, 5} for m = {18, 36} manipulation.