Search papers, labs, and topics across Lattice.
This study introduces LIMMT, a data-centric approach to motion tracking that emphasizes the importance of high-quality motion data in optimizing training trajectories for physics-based humanoid models. By defining motion data quality across three dimensions鈥攑hysics feasibility, diversity, and complexity鈥攖he authors demonstrate that training with as little as 3% of the AMASS dataset can outperform models trained on the full dataset. Extensive experiments confirm that their method not only enhances tracking performance but also improves the efficiency of data usage in motion tracking tasks.
Training with just 3% of high-quality motion data can yield superior tracking performance compared to using the entire dataset.
We argue that high-quality motion data can steer tracking policies toward better optimization trajectories early in training. In this work, we introduce LIMMT (Less Is More for Motion Tracking). To our knowledge, this is the first data-centric study for physics-based humanoid motion tracking. We go beyond simply removing low-quality and erroneous clips, but define motion data quality through three dimensions: physics feasibility, diversity, and complexity. We show that even training with under 3% of AMASS yields better tracking performance than training with the full dataset. We further conduct data cleaning on the estimated web-sourced mocap data. Extensive experiments and analyses validate the effectiveness of our framework.
