@Eric-Katzfey I see. Thanks for the response.

Hi ModalAI team. I’ve been successfully using voxl-tflite-server to run custom neural networks that take image input from the mpa/hires camera. Now, I’d like to expand to more general neural networks that also incorporate other forms of input. For example, prior waypoint information stored in memory or additional sensory data (e.g., IMU, rangefinders, or temporal data from previous frames).

What would be the recommended roadmap to support this?

Specifically:

• How should I modify or extend the current model helper pipeline to feed both image and non-image inputs?

• Is there support in voxl-tflite-server for multiple input tensors (e.g., combining a camera frame other forms of information)?

Hi ModalAI team, I am working on a project that involves developing an end-to-end neural network-based control policy for a quadrotor. For this, I would like to bypass the PX4 autopilot and directly control the drone at the low level using a 4x1 vector of thrust and torque (roll, pitch, yaw) commands, similar to what is used in 12D system models.

My goal is to retain the VOXL infrastructure, especially for reading from onboard services like custom neural network outputs through the voxl-tflite-server pipes and using the existing sensor and system services. However, I would like to disable PX4's control stack and instead send raw actuator commands directly from my control policy.

Could you clarify whether this type of low-level control is supported through VOXL’s existing service architecture? If so, do I need to implement my control policy in C++, or is it also feasible to use Python for sending these thrust and torque commands at a low enough latency for closed-loop flight? Any guidance on how to interface with the motor outputs or appropriate service pipes to do this would be greatly appreciated. Thank you.

@Zachary-Lowell-0 Hi Zach and team, I have been using the voxl docker mavsdk python container successfully for MAVSDK based drone control, and now I would like to read from VOXL’s Modal Pipe Architecture (MPA) inside the same container. I saw your suggestion about either mounting header files like modal_start_stop.h or installing the full MPA package inside the container. Could you please clarify the best way to do this? Should I install the libmodal_pipe Debian package inside the Docker image using the Dockerfile, or manually copy the headers and libraries from the host? Also, is this setup compatible with the voxl docker mavsdk python container, or is it mainly meant for the C++ MAVSDK version?

I am trying to listen to a custom voxl tflite server service that gives inference output from my custom neural networks. I want to deploy some motion planning in Python based on the neural net output. I would really appreciate more specific setup instructions or an example Dockerfile that includes MPA support. Thank you.

@Alex-Kushleyev Thanks for the response.

@Alex-Kushleyev I was having the same problem and thanks to this answer I can build without an issue. But the only thing I am worried is when I deploy this to my Voxl 2 machine with SDK version 1.4.1 and voxl-tflite-server version 0.3.9, would it break anything? My aim is to develop some custom tflite application.

Hi, I have questions about the PX4 Autonomy Developer Kit and Starling 2.

How is the inference time for object detection and classification models, considering that other tasks like flight control are running? Can PyTorch be used for image-based tasks on these systems, or is TFLite the main option? Are there ways to customize compute power, especially GPU acceleration, beyond the built-in Adreno 650?

Hi,

I’m considering purchasing either the PX4 Autonomy Development Kit or the Starling 2 Drone and had a couple of questions about their durability and maintenance:

1- Do these drones come with any kind of warranty? If so, what does it cover?

2- Are potentially breakable parts like the chassis and propellers easily replaceable? I saw a general list of accessories on ModalAI’s website (https://www.modalai.com/collections/all?&sort_by=manual), but I didn’t find anything specific about chassis replacements.

3- Would the Sentinel replacement propellers (https://www.modalai.com/collections/all/products/mrp-d0006-1-01) be compatible with these drones?
I’d appreciate any insights from those who have experience with these models. Thanks!