We introduce a unified approach to forecast the dynamics of human keypoints along with the motion trajectory based on a short sequence of input poses. While many studies address either full-body pose prediction or motion trajectory prediction, only a few attempt to merge them. We propose a motion transformation technique to simultaneously predict full-body pose and trajectory key-points in a global coordinate frame. We utilize an off-the-shelf 3D human pose estimation module, a graph attention network to encode the skeleton structure, and a compact, non-autoregressive transformer suitable for real-time motion prediction for human-robot interaction and human-aware navigation.

Clone the Repository

git clone https://github.com/nisarganc/UPTor.git
cd UPTor

Install Dependencies

python -m venv create ./.uptor  
source ./.uptor/bin/activate  
pip install -r requirements.txt

Download Datasets from here and unzip it to the root of this repository with train and test split folders:

UPTor  
├── darko  
│   ├── test   
|   |  ├── 0001.npy  
|   |  └── ...  
│   ├── train 
|   |   ├── 0017.npy
|   |   └── ...
|   └── darko.yaml
├── cmu_mocap
|   └── ...
└── human_36m
    └── ...

python darko.py
python cmu_mocap.py
python human_36m.py

If you use UPTor code in your research, please cite the following paper

@misc{nilavadi2025uptorunified3dhuman,
  title         = {UPTor: Unified 3D Human Pose Dynamics and Trajectory Prediction for Human-Robot Interaction},
  author        = {Nisarga Nilavadi and Andrey Rudenko and Timm Linder},
  year          = {2025},
  eprint        = {2505.14866},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  url           = {https://arxiv.org/abs/2505.14866},
}