Event
THR: Selecting Detectors by Thermal Headroom under High Temperatures
1 reports · 1 sources
Sparsh Roy and co-authors propose the Thermal-Headroom Routing (THR) model-selection strategy: it uses a closed-form RC thermal model to predict processor temperature and excludes detector configurations projected to cross the throttling threshold. The authors claim that, at ambient temperatures of 45–55°C and under tight inference deadlines, THR eliminated throttling in simulation. The authors' measurements on an Apple M5 support only a single-node temperature-rise approximation and did not verify hardware-level throttling. The authors also note that, at a fixed operating point, a well-tuned thermostat or duty-cycle cap can achieve effects comparable to THR; predictive routing is more useful in scenarios where ambient conditions or calibrated thermal parameters deviate from the values at which the static policy was tuned.
Report timeline
- Thermal-Headroom Routing: Heat-Budget Routing for Robot Perception in Extreme Heat
The authors propose Thermal-Headroom Routing (THR), a lightweight model-selection policy that uses a closed-form RC thermal model to predict processor temperature and exclude detector configurations predicted to cross the throttling threshold.
arXiv cs.RO ·