Search papers, labs, and topics across Lattice.
This paper introduces MotionVLA, a vision-language-action model that addresses the limitations of existing motion generation methods by employing a dual-stream frequency tokenizer (DSFT) to separately compress low-frequency pose and high-frequency velocity signals. By analyzing the frequency-domain characteristics of human motion, the authors demonstrate that traditional single-codebook approaches inadequately represent high-frequency dynamics, leading to biased quantization. Experimental results show that MotionVLA significantly reduces the diversity gap to real data and enhances motion-condition consistency, proving the effectiveness of frequency-aware dual-stream decoupling in autoregressive motion generation.
A dual-stream tokenizer reveals that traditional methods misrepresent high-frequency motion dynamics, leading to over 50% improvement in diversity alignment with real data.
Generating realistic humanoid motion from scene images and text involves both low-frequency pose semantics and high-frequency physical dynamics. However, many existing methods tokenize motion with a single shared codebook, forcing heterogeneous motion signals into the same quantization space. Our frequency-domain analysis of human motion data reveals a clear mismatch between single-codebook quantization and motion statistics: five DCT coefficients capture 93% of joint-position energy but only 37% of joint-velocity energy, which can bias quantization toward pose statistics and under-represent high-frequency velocity components. A second challenge lies in adapting a standard autoregressive model to effectively model high-frequency physical signals in motion sequences. Therefore, we propose DSFT, a dual-stream frequency tokenizer that separates motion into Base and physical streams and compresses them independently with DCT truncation and BPE. Furthermore, we present MotionVLA, a Qwen3.5-based model that arranges Base and physical tokens in a unified sequence, where Phys tokens are predicted after Base tokens. Experiments on HumanML3D and MBench show that, despite using a lightweight 2B backbone, MotionVLA reduces the Diversity gap to real data by over 50% on HumanML3D and improves Motion-Condition Consistency by 3.8% on MBench, supporting frequency-aware dual-stream decoupling as an effective formulation for autoregressive motion generation. Code: https://github.com/AIGeeksGroup/MotionVLA. Website: https://aigeeksgroup.github.io/MotionVLA.
