What is Orin's best practice of software architecture?

Dear community,

I am developing an AGV using Orin AGX. Sensors are three GMSL cameras (30fps 1080p, IMX390), an IMU, a GPS and a 32-channel Lidar. Similar to a typical autonomous driving car, in short I have a neural network as my perception module, a planning module, an EKF module as my state estimator, and a vehicle control module. Currently I am using ROS: running each of the module as a separate ROS node. However, my experience is not very pleasant. First, the delay of GMSL image is around 150ms even with Deepstream( from sensor to perception neural network, including distortion correction). Second, the software system is not robust enough and occasionally crashed during testing. So I am wondering if there are any alternatives for the software architecture? Is ROS2 better?

You may refer to JetRacer to get more idears, see GitHub - NVIDIA-AI-IOT/jetracer: An autonomous AI racecar using NVIDIA Jetson Nano

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.