Research — Huirong Chai

Password required

Diagnosing Interaction Effects in Pose Prediction · 2026

The penalty multi-agent interaction imposes on pose prediction is concentrated below 5 Hz, decays monotonically with frequency, and is largely absent above 10 Hz.

Robots that share space with people must anticipate their partners’ motion. The field evaluates this with a single aggregated error number, which conflates individual motion modelling with interaction modelling and cannot show where interaction-induced error accumulates. A three-prong frequency-domain methodology — Fast Fourier Transform, band-pass filtering, and discrete wavelet transform — separates the two, and all three prongs converge on the same pattern across datasets.

Runtime latency inside an on-device language model can be measured live, phase by phase and kernel by kernel, at under 3% throughput cost.

Running language models locally protects privacy and cuts cost, but the performance-efficiency trade-offs on constrained hardware are largely invisible. LM-Meter is a lightweight online profiler that exposes fine-grained latency during on-device inference without auxiliary equipment, making phase- and kernel-level bottlenecks measurable rather than assumed.