03 Dec

ROS and Keras

Eun Sang 'Jesse' Cha

This blog is a bit technical, but here it goes. The last blog focused on the write up side of my project, but this time, I will focus on how I attempted to replicate my 1/8th scale autonomous RC car results for my 1/32 scale vehicles.

-Data Collection-
The image above is an image straight from the Kogeto Dot Lens, which shows a 360 panoramic view of its environment. This image isn't centered, so an opencv2 function crops the images to only display the information we want. Because if we think about it, the only state information we need is just the donut image around the lens. Data is collected from the Robot Operating System and the Optitrack Motion Capture System, which gets timestamped and sorted out as shown below. We collected around 15 minutes of training data. Everytime a crash event happened, the throttle was maxed to freespin the rear wheels, which was then used as an indicator to crop out the data after the experiment.The data parsing was done by my project teammate Kevin, who will probably blog about it on his blog, and it was based off of the previous parser for the 1/8th scale project.

-Keras in Tensorflow-
The model was generated in Keras, a plug-in for Tensorflow. The images you see below represent the training set and validation set accuracy and loss on TensorBoard. There were a few key insights I obtained while constructing the 3d cnn model for the car, which I will list below. We ended up having an accuracy of 99% on the test set and 98% on the training set, so results were pretty good.

-First Insight: Only Two 3D convolutional layers are sufficient enough to find a relationship between the vehicle images and steering/throttle values. This insight was interesting because I had previously used five 3D.conv layers on my 1/8th scale vehicles.
-Second Insight: 160x160 images creates more accurate inferences than 200x200 images using the same model. This is probably because 200x200 images need deeper networks and 160x160 images sufficiently capture the state of the vehicle.
My next blog will focus on deploying this model and optimizing the model to insure that the vehicle can run fast and robustly. Stay tuned!

Next Post Previous Post

Search

Popular Tags

foray origami robots blimp ducky vr edg SLAM computer vision computational design roco slam Origami rocolib arduino RoCo design Robot Compiler simulation Motion Planning autonomy edg-model mesh reinforcement learning; multi-agent; mean-field; logics create; kalman filter; motors imu friction mesh network tank esp32 networking contacts multi-agent pushing auxetic machine learning dynamics esp now visual processing wireless control communication webapp pcbs structure ROCO i2c path planning cars Robotica reinforcement learning computer skill openmv test motor dataset visual DARPA odometry Battery surface structure power consumption dash port-arrays modeling origami paper frequency diversity semantic infrastructure auxetics feedback gondola machine vision Consensus bistable auxetic controller gestural intuition mixed reality unit cell robot crawler gestural interface open house network create; kalman filter; EM; refactoring 4-wheeled darpa; create; icra2021 apriltags lidar logic control FORAY perception Quantum Information crazyflie 433mhz optimization propeller Foray structures sensors environmental design radio DER Kullback-Leibler Divergence lasercutter tutorial von Mises Fisher Distribution IR simulink legs project management localization distributed optimization blynk Auxetics painlessmesh experiment results offline reinforcement learning exam research proposal altimeter thrust PCB sensor fusion gate or webots multi-agent path planning UAV RL integration test arnhold joystick balloon test vision processing pushing problem modeling contacts mesh networking motor configuration nodemcu swarm object detection fusionbot ros2 rviz2 gazebo conference matlab arduino nano decentralized surfaces keyboard IROS Integrated Demo throughput peripheral device paper review reinforcement learning; motorized boats Writing slam; paper review rhino Short Post range distance surp origami robot nylon pvdf elastic polymers plastic hot plates reinforcement learning; multi-agent; mean-field; game theory; mean field; reinforcement learning current reinforcement learning; multi-agent; Xbee Zigbee robot compiler spi contact neural network dynamic modeling quasi-static modeling edg-examples Kalman Filter port-array von Mises Fisher von Mises-Fisher literature review multipack car Zero Elastic Energy modes swarm control solar cars system ID serial monitor edg-ide Discrete Elastic Rods ceres debug Consensus Algorithms coriolis force dynamic model proposal wtf deep learning pdms interpolation wireless signal strength reinforcement learning; multi-agent; decentralized misc reinforcement learning; multi-agent; mean-field; stationary demo video flytrap papers steering structured field pivoting differential power detectionrate color shape edg-refactoring pain python vision Machine Learning Q-Learning Model-Free release conferences quasi-static pushing liquid crystal elastomer nonholonomic systems iot grant feedback-controlled controls trajectory optimization surface contact Characterization tool ML multi-agent path finding Liquid Crystal Elastomers Learning kinematic_model reservoir computing 2.4 GHz concept 433 MHz Bayesian Network creative Computer skill internal silicone integrated system instance segmentation marginal cost pricing bootstrapping Bayesian networks ensemble methods offline Rl hardware manipulation COVID 19 paper cutter ESP32 kalman filter; multi-robot state space delta tolling buoyancy chiral square unit cell pressure distribution model prediction thrust configuration grid fabrication firmware protobuf #activecapturingmechanism #claw 3D-CHESS; kalman filter boxboat tracking model-based RL von Mises Distribution slam; flowchart; spring-mass 3D-CHESS; 433 rssi signal strength multirobot system writing in-situ design buoyancy equations boat deployment cell cameras der newton's method Barycenter Euler-Bernoulli beam sma-based actuator game theory Bessel functions von Mises-Fisher Distributions ide HDL board design battery current sensor ZMP center of friction Multi-Agent Systems code balloon semantic segmentation MIMO communications ICRA ICML guassian processes probabilistic neural networks graph generators MBRL batch learning ros edg-compiler multirobot optitrack adaptive control motor driver submission DNN kalman filtering point cloud 2.4ghz latency scenic codesign presentations machine training preliminary exam multiple devices inkscape material multi-object tracking ICLR Python actuation catching mechanism integration flapping color_segmentation lab_data result segmentation on multiple object and centers finding competition prelab multiple blimps pid defender Synthesizable Radio 3d printing kalman filter motor distribution barometer detection lift force depth estimation cad solidworks design tool pixracer pixhawk ardupilot flight controller regression fly test sensor selection