[tridge]

Summary

• use TMPDIR, then TMP, for survey runtime status so MT11 camera-app and web processes share a writable path
• reject external MAVLink, SIYI, UniGCS, and web pointing while Survey owns the gimbal, while preserving internal survey steering
• hand control back to pilot or GCS pointing after ownership takeover pauses the survey, and report refused claims while the mount is retracted
• support MAV_CMD_CAMERA_TRACK_POINT using the existing rectangle tracker and advertise point-tracking capability
• expand unit, SITL, and documentation coverage for the new behavior

Validation

• native ARM camera-app and SITL builds passed during survey validation
• manual-control, UniGCS time, survey lifecycle, MAVLink time, and TMP-path tests passed
• expanded survey-control SITL coverage passed, including pointing guards, ownership takeover, mode exits, and web lease conflicts
• moving-flight survey SITL passed at 18 m/s for 45 seconds with 40 unique captures and stream/SD pixel parity
• point-tracking unit and SITL regression coverage is included in this branch

No additional tests were run solely while opening this PR.

[AP-Review]

Deprecated — see below for the updated review.

Previous review (2026-10-19)

Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting.
Verdict: REQUEST CHANGES

Reviewed at head 83de1b111d.
Full report: https://firmware.ardupilot.org/Tools/APReview/DevCallReviews/PRReviews/ardupilot/ap_cameragimbal/54/1.html#prAP_CameraGimbal-54

Thanks for this. I found one blocking issue in the ownership handover. When the survey is granted the gimbal, survey_clear_external() (https://github.com/ardupilot/ap_cameragimbal/pull/54/files#diff-3dbc5ed84dceb7214a2a88f740c9ce05367c9f98ffe4d140ec2965d47833cd90R82) calls stop_tracking_rate(), but that only sends a zero rate if the camera's own ROI controller produced the rate. A GIMBAL_DEVICE_SET_ATTITUDE rate the autopilot sent while the survey was waiting is not cancelled. On MT11 and the SIYI A8/ZR10 backends that rate stays latched, and the new guard (https://github.com/ardupilot/ap_cameragimbal/pull/54/files#diff-862008e059849dbb1a8cdab3af55eb729c57bbfe996836b8f379ad2c54d054bbR2177) now rejects every external stop. The survey sends nothing of its own until it has terrain/telemetry and straight flight. The companion ArduPilot branch grants the claim while a GCS rate target is active and stops sending targets, so this can happen in practice. Sending an unconditional zero rate in survey_clear_external() would fix it.

Smaller points: docs/survey-mode.md (https://github.com/ardupilot/ap_cameragimbal/pull/54/files#diff-dc698906e8df9372d9cfa691c0b660ed2add2f3811f6cacf62c69b6ba557e10cR86) says a waiting survey keeps claiming through external pointing. In fact SIYI/UniGCS motion, web PTZ, ROI LOCATION and image tracking started while it waits all pause it via ca_mavlink_server_suspend_gimbal() (https://github.com/ardupilot/ap_cameragimbal/pull/54/files#diff-862008e059849dbb1a8cdab3af55eb729c57bbfe996836b8f379ad2c54d054bbR2083), the ROI handler and update_survey(), and a later grant does not resume it. Either the docs or the code should change. Also, ca_mavlink_server_survey_mode() (https://github.com/ardupilot/ap_cameragimbal/pull/54/files#diff-eb0f28c2ab056405d068f7d7a13cdeebbee275550dbfd07943d55015d3caef42R34) no longer has any callers.

Not covered here: the Plane integration test, SITL for targets other than MT11, real gimbal hardware, and a companion-ArduPilot runtime.

[AP-Review]

Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting.
Verdict: COMMENT

Reviewed at head 6efbc1740c.
Full report: https://firmware.ardupilot.org/Tools/APReview/DevCallReviews/PRReviews/ardupilot/ap_cameragimbal/54/2.html#prAP_CameraGimbal-54

No blockers remain at 6efbc17; one minor error-path point is left.

Since the last review, the latched-rate handover blocker is fixed: survey_clear_external() now always sends a zero rate. The waiting-survey docs now match the code, and the unused ca_mavlink_server_survey_mode() is gone.

Minor (1)

• camera_app/src/protocol/survey_server.inc:85: if this zero-rate write fails, nothing retries it, because owned is set anyway and later grants skip the clear, so a direct FC rate can keep driving the gimbal. Keep a pending stop and retry it, as stop_motion() in manual_control.cpp does.

Not fully reviewed: the Plane integration test, SITL for targets other than MT11, release builds, real gimbal hardware and a companion-ArduPilot runtime were not exercised.