Social Motion Prediction with Cognitive Hierarchies

Authors: Wentao Zhu, Jason Qin, Yuke Lou, Hang Ye, Xiaoxuan Ma, Hai Ci, Yizhou Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct comprehensive experiments to validate the effectiveness of our proposed dataset and approach.
Researcher Affiliation Academia 1 Center on Frontiers of Computing Studies, School of Computer Science, Peking University 2 Institute for Artificial Intelligence, Peking University
Pseudocode No The paper describes the model architecture and training objectives mathematically and textually, but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The paper states: 'We plan to release the processed 3D human motion sequences under a license agreement that permits their use for non-commercial scientific research purposes.' This refers to the dataset, not the source code for the methodology or experiments. No other explicit statement or link for the code is provided.
Open Datasets Yes We present Wusi, the first large-scale multi-human 3D motion dataset featuring intense and strategic interactions. We conduct experiments with the following baseline methods including a naive baseline and multiple state-of-the-art approaches: ... 0 20 40 60 80 100 mm Frozen HRI MRT Figure 4: Comparison of baseline methods on CMU-Mocap [1] and our dataset. The baseline methods are trained and tested on two datasets separately. [1] Cmu graphics lab motion capture database. http://mocap.cs.cmu.edu/.
Dataset Splits No The paper mentions training and testing on datasets but does not explicitly specify train/validation/test splits, percentages, or sample counts for reproducibility.
Hardware Specification Yes We implement the proposed framework with Py Torch [47] using a Linux machine with 1 NVIDIA V100 GPU.
Software Dependencies No The paper mentions using 'Py Torch [47]' and 'Adam [33] optimizer' but does not specify their version numbers or other ancillary software with versions required for reproducibility.
Experiment Setup Yes We implement the presented framework to train and test on the proposed Wusi dataset. We employ Transformer encoder [62] for both the local and global state encoders, as well as Transformer decoders for the policy networks. Each Transformer consists of 3 layers with 8 attention heads. We share parameters for policy networks φ(1) . . . φ(K). We set the strategic reasoning depth K = 3 unless otherwise stated. For all the methods, we provide 1s motion history and predict future 1s motion... We set λ = 0.002, batch size 32, learning rate 0.0001, and train for 60 epochs using Adam [33] optimizer.