Stochastic Prediction of Multi-Agent Interactions from Partial Observations

Authors: Chen Sun, Per Karlsson, Jiajun Wu, Joshua B Tenenbaum, Kevin Murphy

ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show that our method outperforms various baselines on two sports datasets, one based on real basketball trajectories, and one generated by a soccer game engine.
Researcher Affiliation Collaboration Chen Sun Google Research Per Karlsson Google Research Jiajun Wu MIT CSAIL Joshua B Tenenbaum MIT CSAIL Kevin Murphy Google Research
Pseudocode No The paper does not contain any section explicitly labeled 'Pseudocode' or 'Algorithm', nor are there structured steps formatted like an algorithm block.
Open Source Code No We plan to release the videos along with the game engine after publication of the paper. Video samples can be found at bit.ly/2E3qg6F
Open Datasets Yes We use the basketball data from Zhan et al. (2018).
Dataset Splits Yes The hyper parameters, such as the base learning rate and the KL divergence weight β, are tuned on a hold-out validation set.
Hardware Specification Yes The models are trained on 6 V100 GPUs with synchronous training with batch size of 8 per GPU, we train the model for 80K steps on soccer and 40K steps on basketball.
Software Dependencies No The paper mentions software components such as Res Net-18, S3D, GRUs, MLPs, relation networks, and the Unity game engine, but it does not provide specific version numbers for any of these components or other libraries.
Experiment Setup Yes Our visual encoder is based on Res Net-18 (He et al., 2016), we use the first two blocks of Res Net to maintain spatial resolution, and then aggregate the feature map with max pooling. The encoder is pre-trained on visible players, and then fine-tuned for each baseline. For the soccer data, we down-sample the video to 4 FPS, and treat 4 frames (1 second) as one step. We consider 10 steps in total, 6 observed, 4 unobserved. We set the size of GRU hidden states to 128 for all baselines. The state decoder is a 2-layer MLP. For basketball data, we set every 5 frames as one step, and consider 10 steps as well. The size of GRU hidden states is set to 128. The models are trained on 6 V100 GPUs with synchronous training with batch size of 8 per GPU, we train the model for 80K steps on soccer and 40K steps on basketball. We use a linear learning rate warmup schedule for the first 1K steps, followed by a cosine learning rate schedule. The hyper parameters, such as the base learning rate and the KL divergence weight β, are tuned on a hold-out validation set.