Georgia State University

Robots that share our space have to anticipate us. I measure whether they can.

Two lines of work: how the field evaluates multi-agent motion prediction for human–robot teaming, and what language models actually cost at runtime on the devices people carry.

Research Publications Get in touch

FieldHuman–Computer Interaction

FocusHuman–robot teaming · On-device ML

TrackingCOCO-17 · forecast t+3

ObservedSensing platform

LiveOn-device inference · phase latency

Illustrative trace. Real measurement is what the research is for.

What I study

01On-device language model systemsMaking the runtime cost of running language models on phones and edge hardware visible, so it can be optimised rather than guessed at.
02Human–robot teaming and motion predictionUnderstanding how well systems anticipate people’s movement when robots and people share a space.

Selected work

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.
Unlock to read
LM-Meter · 2025
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.
Read the paper

Selected publications

2025LM-Meter: Unveiling Runtime Inference Latency for On-Device Language ModelsSEC

All publications