Section 55.3: Edge vs. cloud-robot computation; asynchronous inference

For Edge vs. cloud-robot computation; asynchronous inference, deployment quality is measured by the command stream, safety monitor state, and replayable evidence behind each command.

A Careful Control Loop

Problem First

Cloud inference can add model capacity, but network loss, privacy constraints, and variable latency can break closed-loop control.

The practical question is therefore specific: which observation arrives, which state estimate is trusted, which action is allowed, which monitor can interrupt it, and which artifact proves the claim afterward?

Theory

Compute placement is a control allocation problem. Safety-critical loops must remain local, while cloud services are appropriate only when latency, privacy, and availability constraints still preserve the action contract.

A compact rule is to keep a computation local when it is both action critical and deadline sensitive. One simple score is

$$R = w_1 \cdot \text{criticality} + w_2 \cdot \text{latency sensitivity} + w_3 \cdot \text{availability risk} + w_4 \cdot \text{privacy cost}.$$

High-$R$ functions belong on the robot or on a trusted edge node. Low-$R$ functions, such as batch summarization, fleet analytics, or nonurgent semantic search, can move to the cloud.

Worked Example

A home robot may safely ask the cloud to summarize a room inventory, but it may not rely on the cloud to decide whether to brake before contacting a person or a wall. That distinction is architectural, not cosmetic.

tasks = {
    "emergency_stop": {"criticality": 5, "latency": 5, "availability": 5, "privacy": 2},
    "global_semantic_search": {"criticality": 1, "latency": 2, "availability": 2, "privacy": 3},
    "task_replanning": {"criticality": 3, "latency": 3, "availability": 3, "privacy": 2},
}

def placement_score(v):
    return v["criticality"] + v["latency"] + v["availability"] + v["privacy"]

placement = {
    name: ("local_or_edge" if placement_score(vals) >= 12  # score >= 12 means high combined criticality, latency, and availability demand; must stay local
           else "cloud_ok")
    for name, vals in tasks.items()
}
print(placement)

{'emergency_stop': 'local_or_edge', 'global_semantic_search': 'cloud_ok', 'task_replanning': 'local_or_edge'}

The expected output should separate what must remain available during network loss from what can tolerate delay. If a remote service appears in the local-or-edge set, the design implication is immediate: the system needs a colocated implementation or a safe degraded fallback.

Practical Recipe

1. Write the observation, action, monitor, metric, and artifact fields before selecting a model.
2. Run a deterministic smoke test and one named perturbation from the panel.
3. Log success, safety events, latency, energy or resource use, and recovery status in the same row group.
4. Compare only methods evaluated by the same script on the same panel and seed plan.
5. Attach a short postmortem to each failed rollout so the artifact remains useful after the plot is forgotten.

Edge vs. cloud-robot computation; asynchronous inference becomes operational when the metric is tied to a runtime interface. The interface names the sensor stream, state estimate, action representation, timing budget, safety or robustness monitor, and deployment artifact.

The disciplined habit is to separate three claims. The conceptual claim explains why the method should help. The systems claim explains which interface it changes. The evidence claim records which measurement would convince a skeptical builder.

Cross-References

For Edge vs. cloud-robot computation; asynchronous inference, connect benchmark design, sim-to-real transfer, uncertainty, and safety barriers through the deployment artifact that will be checked before release.

When a cloud-edge architecture fails, classify the failure as uplink loss, stale cache, remote timeout, inconsistent model versions, bandwidth collapse, or unsafe fallback routing. Then replay the exact sequence with the network behavior fixed at one perturbation setting.