This article is about machine manufacturing for Maritime RobotX Challenge 2018. In a word, we lacked everything: number of developers, period for developing, experience, money, place, and so on. We began designing machines in April 2018 (last only 7 months for shipping) with only four students including three seniors who had to do graduation […]
On May 9th we met Hiroshi Ohta, the CEO of Cloudian K.K. at Osaka Prefecture University. Cloudian is our GOLD sponsor and supports our activities to participate in Martime RobotX Challenge. We already have been provided three AI Boxes from Cloudian. The AI Box has a GPU which performs high speed processing, and an LTE/Wi-Fi […]
We are using VMRC simulator for testing our control system. We build simple PI control system. Inside PID Control subsystem is below. We built Matlab/Simulink model for wam_v control system and convert it to C++ ROS node.
Recently, VMRC simulation package was uploaded to this URL. We succeeded to build this software and write Japanese Documentation about this package
Yesterday, we built opencv-3.3.1 and contrib for Jetson TX2. After opencv-3.3, opencv contains yolo deep neural network. So, we make a simple ROS node for object detection by using tiny yolo and try it!! The code is here. This node can run only 2Hz. This is 5 times as slow as using using darknet in […]
ros-kinetic-opencv3 package is very useful for ROS users, but this package is not supported cuda. So, we built ros-kinetic-opencv3 packages with cuda modules in Jetson TX2. Compiled binary is here. Usage is below. 1.install dependencies 2.remove ros-kinetic-opencv3 3.install dpkg We built this .deb file in this way.(<a href=”https://qiita.com/kendemu/items/a805b0b9828b6f6031db”>reference</a>) 1.get ros-kinetic-opencv3 source from apt-get 2.remove source files […]
In order to use Jetson TX2 and Deep Learning in this competition, I tried to run darknet in Jetson TX2 and tested Jetson TX2 throughput. In this post, I used Tiny-Yolo deep neural network in Jetson TX2. This is the result of object recognition. Tiny yolo structure is here. dog.jpg eagle.jpg giraffe.jpg person.jpg scream.jpg Prediction […]
We use tensorflow in order to use deep learning algorithms such as Faster-RCNN, Yolo, VoxelChain. Compiled .whl file is here. If you want to install tensorflow into Jetson TX2, you should follow these instructions. 1. download jetson_tensorflow.zip 2. unzip jetson_tensorflow.zip 3. cd jetson_tensorflow 4. sh install.sh We checked this .whl file by using this python […]
We build object recognition system using point cloud data. This node are composed of ros_ship_recognition package. The object_recognition_node detect objects by using SHOT features. We conducted a presentation about this node in ROS Japan user group meeting in Kansai. 3D Object Detection using PCL (Point Cloud Library) from masaya kataoka Demonstration movie is here.