Fit the hardware
Profile policy execution against the target device’s memory and compute capacity.
03 / ON-ROBOT INFERENCE
Run the policy within the robot’s limits.
Quantize and optimize inference for the target accelerator, balancing latency, memory, throughput, and task performance.
Talk to us about FieldEdge ↗
THE EDGE DEPLOYMENT LAYER03 — INFERENCE
Meet the control loop
Measure end-to-end inference latency and throughput on the robot’s onboard computer.
Check task performance
Evaluate the optimized policy on real tasks to track the effect of quantization and runtime changes.