@Matt-Turi how would I manually start /run/mpa/tflite_data? I switched to a different m500 and started voxl-tflite-server, but there appears to be no /run/mpa/tflite_data pipe, only the /run/mpa/tflite pipe. The model I'm using is the default MobileNetV2 object detection model for VOXL, and I reset the drone to factory defaults using voxl-configure-tflite

@Matt-Turi Yep, turns out switching to wall power fixed it. My battery power was too low. Thanks!

@Matt-Turi that did reinstall voxl-tflite-server 0.1.8, but still having the abort issue.
voxl-version gives the same output, and voxl-inspect-services -v says the version of voxl-tflite-server is 0.1.8 as expected

@Matt-Turi I downloaded voxl-tflite-server_0.1.8.ipk from http://voxl-packages.modalai.com/stable/, pushed it to the drone via adb, and ran opkg install voxl-tflite-server_0.1.8.ipk. However, it seems that 0.1.8 is already installed:

 * opkg_prepare_file_for_install: Refusing to load file 'voxl-tflite-server_0.1.8.ipk' as it matches the installed version of voxl-tflite-server (0.1.8).

I tried installing version 0.2.0 from dev instead, but it requires upgrading libmodal_pipe, libmodal_json, and voxl-opencv and I don't really want to do an upgrade right now in fear of breaking things

@Matt-Turi I actually don't have a custom version running, so I don't know how that happened. I'll reinstall voxl-tflite-server from the voxl packages page and see if that fixes things

@Matt-Turi
Here's the output of voxl-version

cat: /etc/modalai/voxl-software-bundle-version.txt: No such file or directory
--------------------------------------------------------------------------------
system-image:    ModalAI 3.3.0 BUILDER: ekatzfey BUILD_TIME: 2021-06-06_19:28
kernel:          #1 SMP PREEMPT Sun Jun 6 19:41:01 UTC 2021 3.18.71-perf
factory-bundle:  1.0.1 (Yocto installation)
--------------------------------------------------------------------------------
architecture:    aarch64
processor:       apq8096
os:              GNU/Linux
--------------------------------------------------------------------------------
voxl-suite:
Package: voxl-suite
Version: 0.5.0
Depends: libjpeg_turbo (>= 9.0.4), libmodal_exposure (>= 0.0.2), libmodal_json (>= 0.3.6), libmodal_pipe (>= 2.1.1), librc_math (>= 1.1.5), libvoxl_cutils (>= 0.0.2), libvoxl_io (>= 0.5.4), mavlink-camera-manager (>= 0.0.2), mongoose (>= 7.3.0), opencv (>= 4.5.2-2), openmp (>= 10.0.2), voxl-camera-calibration (>= 0.1.1), voxl-camera-server (>= 0.8.1), voxl-cpu-monitor (>= 0.2.0), voxl-dfs-server (>= 0.2.2), voxl-docker-support (>= 1.1.3), voxl-gphoto2 (>= 0.0.5), voxl-imu-server (>= 0.9.1), voxl-mavlink (>= 0.0.2), voxl-modem (>= 0.12.0), voxl-mpa-tools (>= 0.3.6), voxl-nodes (>= 0.2.0), voxl-portal (>= 0.1.2), voxl-qvio-server (>= 0.3.4), voxl-streamer (>= 0.2.6), voxl-tag-detector (>= 0.0.2), voxl-tflite (>= 2.2.3), voxl-tflite-server (>= 0.1.5), voxl-utils (>= 0.8.4), voxl-vision-px4 (>= 0.9.5), voxl-vpn (>= 0.0.3)
Status: install user installed
Architecture: all
Installed-Time: 2166

Here's the output of `voxl-inspect-services -v"

 Service Name          |  Version  |  Enabled  |   Running   |  CPU Usage
----------------------------------------------------------------------------
 docker-autorun        |    1.1.3  |  Enabled  | Not Running | Not Running
 docker-daemon         |    1.1.3  |  Enabled  |   Running   |     1.0
 modallink-relink      |   0.12.0  | Disabled  | Not Running | Not Running
 voxl-camera-server    |    0.8.1  |  Enabled  |   Running   |     7.1
 voxl-cpu-monitor      |    0.2.0  |  Enabled  |   Running   |     0.0
 voxl-dfs-server       |    0.2.2  |  Enabled  |   Running   |     8.0
 voxl-drogon-detector  |    0.3.4  | Disabled  | Not Running | Not Running
 voxl-imu-server       |    0.9.1  |  Enabled  |   Running   |     4.4
 voxl-modem            |   0.12.0  | Disabled  | Not Running | Not Running
 voxl-portal           |    0.1.2  |  Enabled  | Not Running | Not Running
 voxl-qvio-server      |    0.3.4  |  Enabled  |   Running   |     0.0
 voxl-streamer         |    0.2.6  | Disabled  | Not Running | Not Running
 voxl-tag-detector     |    0.0.2  |  Enabled  | Not Running | Not Running
 voxl-tflite-server    |    0.8.5  |  Enabled  | Not Running | Not Running
 voxl-time-sync        |    0.9.5  | Disabled  | Not Running | Not Running
 voxl-vision-px4       |    0.8.5  |  Enabled  |   Running   |     4.0
 voxl-wait-for-fs      |           |  Enabled  |  Completed  |  Completed

I'm trying to run voxl-tflite-server on an m500, but when I run the command, this is the output:

=================================================================
skip_n_frames:                    5
=================================================================
=================================================================
model:                            /usr/bin/dnn/ssdlite_mobilenet_v2_coco.tflite
=================================================================
=================================================================
input_pipe:                       /run/mpa/hires/
=================================================================

------VOXL TFLite Server------

Aborted

I tried running voxl-configure-tflite again, but same result. My configuration file from /etc/modalai/voxl-tflite-server.conf looks like this:

/**
 * This file contains configuration that's specific to voxl-tflite-server.
 *
 * skip_n_frames - how many frames to skip between processed frames. For 30hz
 *                   input frame rate, we recommend skipping 5 frame resulting
 *                   in 5hz model output.
 * model         - which model to use. Currently support mobilenet for
 *                   object detection or midas for monocular depth
 * input_pipe    - which camera to use (tracking or hires).
 */
{
        "skip_n_frames":        5,
        "model":        "/usr/bin/dnn/ssdlite_mobilenet_v2_coco.tflite",
        "input_pipe":   "/run/mpa/hires/"

Actually, using docker import fixed it, nvm!

I'm trying to build a customized version of the roskinetic docker image mentioned in the docs. I added opencv, cv_bridge, and other misc dependencies to the roskinetic-xenial docker image by installing them with apt inside the image, committing the changes to a new image (roskinetic-xenial:opencv), running docker save roskinetic-xenial:opencv > roskinetic-xenial-opencv.tar, and running adb push roskinetic-xenial-opencv /data/. I get this error when I run docker load -i roskinetic-xenial-opencv.tar inside /data/:

Error response from daemon: open /data/tmp/docker-import-737782374/repo/a36ddef81cb09cd9fa7749e7d29fe7ff252301cb330d369f05648c681e726448/layer.tar: no such file or directory.

I tried deleting the contents of /data/tmp and restarting VOXL to no avail. I was able to get the original roskinetic-xenial:v1.0 docker image running following the docs linked above, but for some reason, this customized docker image isn't loading.

Thanks! I'll try this out