Lens 04: Sensitivity Surface. Which knob matters most?

WiFi latency WAS the one parameter with a cliff edge — for a long time, the most important knob in the entire system. But this lens now describes a split world, not a unified one.

Below 30 milliseconds, the navigation loop runs cleanly. VLM inference takes 18 milliseconds, the command round-trip adds another 15, and the total cycle stays well under 50 milliseconds. Between 30 and 80 milliseconds there is degradation, but it is recoverable: the EMA filter absorbs jitter, the robot slows slightly, and collisions remain rare.

Then, at approximately 100 milliseconds, the system crosses a discontinuity. At one meter per second, 100 milliseconds of WiFi latency adds 10 centimeters of positional uncertainty per command. Three or four stacked spikes push the total loop delay past 150 milliseconds, long enough for a chair leg to appear between when the VLM saw clear space and when the motor actually fires.

This is where the new finding changes the picture. Annie's Pi 5 carries a Hailo-8 AI HAT Plus — a 26 TOPS neural accelerator that has been sitting unused for navigation. Activating it gives the safety layer a WiFi-independent path: YOLOv8 Nano runs locally at 430 frames per second with under 10 milliseconds latency, producing pixel-precise obstacle bounding boxes without a single packet traversing the network.

The IROS paper at arXiv 2601.21506 validates this split experimentally. A fast local System 1 paired with a slow remote System 2 cuts end-to-end latency by 66 percent and lifts task success from 5.83 percent to 67.5 percent. With Hailo-8 active, obstacle avoidance no longer depends on WiFi at all. The bar for the safety path drops from 95 percent cliff-edge coral to 15 percent green — a forgiving parameter instead of a catastrophic one.

The cliff edge still exists — but only for the semantic path. "Where is the kitchen?" "What room is this?" "Is the path blocked by a glass door?" These queries require open-vocabulary VLM reasoning on Panda, and they will always traverse WiFi. But they are never the thing that lets a chair leg hit the chassis. The knob that could kill the robot has been converted into a knob that can merely slow its higher cognition.

The second catastrophically sensitive parameter is motor speed for turn commands. At motor speed 30, a 5-degree turn request produces 37 degrees of actual rotation — a 640 percent overshoot driven by momentum. The transition between controllable and oscillating behavior is sharp, not gradual.

The most surprising finding about VLM frame rate above 15 Hertz is how insensitive it is. At one meter per second, two frames captured one-fifteenth of a second apart differ by only 6.7 centimeters of robot travel. The multi-query pipeline's value is not speed. It is diversity. Spending alternate frames on scene classification, obstacle description, and path assessment costs nothing in navigation responsiveness while tripling semantic richness.

EMA alpha at 0.3 sits in the medium band — important, but with a wide optimum and no cliff edge.

The bottom line has changed. Before: fix WiFi before touching anything else. Now: activate Hailo-8 before touching anything else. It removes the only failure mode where a WiFi glitch can cause a physical collision, and it costs nothing in new hardware. The WiFi channel itself is still worth optimizing — dedicated 5-gigahertz, wired Ethernet bridge — but it becomes a UX optimization, not a safety prerequisite. Annie is a dual-process robot now. Reflexes on the Pi. Reasoning on Panda. The cliff edge on the semantic path is a latency problem, not a safety problem.