pub struct TierInputs {
pub has_accelerator: bool,
pub local_inference_capable: bool,
pub models_fit_locally: bool,
pub compute_node_paired: bool,
pub bearer_acceptable: bool,
pub can_run_light_local: bool,
}Expand description
The inputs the agent reads to pick a perception tier.
Fields§
§has_accelerator: boolThe board has an NPU / GPU.
local_inference_capable: boolThe board has no NPU but a CPU strong enough to run the detector locally
(for example a Cortex-A76-class SoC). A distinct signal from
has_accelerator: it is a full local compute path, so it resolves to
Local the same way an accelerator does, and the board runs detection
on-board rather than offloading. Set it only when the board really has
that path, so the picker never invents one.
models_fit_locally: boolThe required models fit + run on the local accelerator.
compute_node_paired: boolA compute node is paired (local-first, over the LAN or a relay).
bearer_acceptable: boolThe bearer’s latency is inside the offload budget for the mission.
can_run_light_local: boolThe board can run a light local detector + fast VIO (the control-loop path) even though it cannot run the heavy models.
Implementations§
Source§impl TierInputs
impl TierInputs
Sourcepub fn for_drone(
has_accelerator: bool,
local_inference_capable: bool,
compute_node_paired: bool,
bearer_acceptable: bool,
) -> Self
pub fn for_drone( has_accelerator: bool, local_inference_capable: bool, compute_node_paired: bool, bearer_acceptable: bool, ) -> Self
The tier inputs for a drone: the local compute picture (an NPU, or a CPU
strong enough for the detector) plus the offload-link signal (a paired,
reachable compute node). models_fit_locally is held true for an
accelerator board (assumed to fit its recommended detector, as there is
no model-fit probe) and can_run_light_local false (no split
light-detector plus fast-VIO path). Build every status view through this
so they all feed pick_tier identically.