WEBVTT 00:00:00.000 --> 00:00:06.120 A vision processor gives the program an observation, not a motor command. 00:00:06.120 --> 00:00:16.380 Make the decision after the observation explicit: pass every gate, or return UNKNOWN with a reason and fallback. 00:00:16.440 --> 00:00:27.960 A region processor and an AprilTag processor do not expose the same measurements or one universal confidence number. 00:00:27.960 --> 00:00:36.360 A region result can check its strongest score, margin over the second-best score, and stability. 00:00:36.480 --> 00:00:51.420 An AprilTag result can check for a detection and metadata, an allowed ID, reviewed range and bearing, and freshness. 00:00:51.420 --> 00:01:04.140 It can require the same ID for three accepted frames. The range, bearing, age, and score limits are examples. 00:01:04.140 --> 00:01:17.280 The team must choose and validate its real policy from labelled evidence such as tag size and reprojection error. 00:01:17.280 --> 00:01:28.080 Every gate should be visible, and every failure should have a name. 00:01:28.080 --> 00:01:38.220 The Decision object keeps together the observed result, the autonomous branch, and the acceptance or rejection reason. 00:01:38.220 --> 00:01:48.540 The region policy rejects a weak winner, a low-margin winner, or an unstable result. 00:01:48.540 --> 00:01:59.280 The AprilTag policy rejects no detection, missing metadata, or a disallowed tag ID before later gates can pass. 00:01:59.280 --> 00:02:10.920 After those checks pass, it counts consecutive frames. The first two return UNSTABLE; the third may select a branch. 00:02:10.920 --> 00:02:23.400 Any failed gate or ID change resets the count. The FTC API supplies detections and pose fields. 00:02:23.400 --> 00:02:33.720 Observation age and the limits shown here are team-defined policy, not a universal SDK confidence field. 00:02:33.720 --> 00:02:44.520 Now switch back to the overall view and name each possible outcome. 00:02:44.520 --> 00:02:55.200 A weak region becomes TOO_WEAK. Two close regions become LOW_MARGIN. 00:02:55.200 --> 00:03:05.280 AprilTag failures include NO_DETECTION, MISSING_METADATA, DISALLOWED_ID, OUT_OF_BOUNDS, and STALE. 00:03:05.280 --> 00:03:14.640 A candidate that has not survived enough frames is UNSTABLE. 00:03:14.760 --> 00:03:21.900 Only an observation that passes the whole policy with no failed gates becomes ACCEPTED. 00:03:21.900 --> 00:03:35.880 Named reasons show which assumption failed and prevent an old accepted branch from surviving after a new rejection. 00:03:36.540 --> 00:03:49.080 The branch table stays small. An accepted observation selects left, center, or right. 00:03:49.080 --> 00:03:55.920 Anything rejected produces UNKNOWN and selects CENTER_SAFE in this example. 00:03:55.920 --> 00:04:09.720 CENTER_SAFE is not proof that center is safe every season. It only names the fallback in this example fixture. 00:04:09.720 --> 00:04:15.240 The team must choose its real fallback. 00:04:15.240 --> 00:04:21.960 Replay labelled observations with robot outputs held at zero. 00:04:21.960 --> 00:04:38.280 Record the first failed gate, result, reason, selected branch, and proof that rejected data never requests motion. 00:04:38.280 --> 00:04:48.780 Only after the policy passes should the actual camera and robot be validated with proper precautions. 00:04:48.780 --> 00:05:06.360 Reliable vision requires many checks, testing, and tuning. Try the policy against real evidence and keep improving it.