What it is
Onboard compute links model ambition to latency, power, thermal design, memory, and autonomy.
Why it matters
Robots cannot always wait for remote inference; contact-rich tasks need fast local loops.
How not to overread it
TOPS or FLOPS alone do not predict deployed robot behavior.
Related edges
Robot policy
Onboard inference
Compute spec does not imply task success.
Inference latency
Real-time operation
Tail latency and jitter matter.
Energy and thermal budget
Robot hardware design
Chip-level numbers need system context.
Physical AI
Robot deployment
Physical AI claims need power, latency, and thermal context.