Latency budget
The total time allowed from sensor measurement to actuator command in a sense-plan-act loop. Each stage of the publish-subscribe pipeline (sensor capture, driver processing, middleware delivery, callback execution, planning, command delivery) consumes part of this budget. Exceeding the latency budget at speed produces unsafe robot behavior.
Defined in 2 GAGE programs, which carry 4 distinct definitions of it. The wording above is taught in Foundations of Robotics and Physical AI.
How each discipline defines it
The same term does different work depending on who is using it. These are the definitions as each program teaches them, unedited.
The total time allowed from sensor measurement to actuator command in a sense-plan-act loop. Each stage of the publish-subscribe pipeline (sensor capture, driver processing, middleware delivery, callback execution, planning, command delivery) consumes part of this budget. Exceeding the latency budget at speed produces unsafe robot behavior.
The maximum acceptable time a content moderation check may add to a generation request before user experience is meaningfully degraded, allocated across prompt-side, output-side, and, for streaming products, incremental output-side filtering.
The maximum allowable time for a function or loop cycle to complete, determined by the physics of the control task. For a 1 kHz controller, the latency budget is 1 millisecond. Exceeding the budget even occasionally violates control stability assumptions.
The total time allocated to a single control loop iteration, broken into fractions for OS wakeup latency, control law computation, and sensor/actuator communication. Designing a latency budget before writing code ensures the chosen control rate is feasible on the target hardware and prevents discovering the infeasibility after assembly.
Terms it appears with
Not an alphabetical neighbourhood: these are the terms taught in the same lessons, ranked by how often they appear together.