Edge2AI Autonomous Car
Overview
NOTICE
As of January 31, 2021, this tutorial references legacy products that no longer represent Cloudera’s current product offerings.
Please visit recommended tutorials:
- How to Create a CDP Private Cloud Base Development Cluster
- All Cloudera Data Platform (CDP) related tutorials
Generating Training Data
The Cloudera Self Driving Vehicle (CSDV) is a miniature autonomous car which is trained to follow markers along a race track, the team achieved this by rapidly sampling photographs and using them to run inference to adjust the steering angle of the vehicle.
The point of the exercise is to demonstrate the Edge to AI life cycle for fast model deployment to the Jetson TX2 module aboard the miniature car. The car takes advantage of Cloudera Edge Manager (CEM) to continue gathering data and improve the model.
The clip below is an example of CSDV driving autonomously around it's racetrack
Introduction
The variety of edge devices--whether it be IoT devices, Cloud VMs, or even containers--generating data in today's industry continues to diversify and can lead to data being lost. There is a need to author flows across all variety of edge devices running across an organization; further, there is a need to monitor the published across all devices without writing customized applications for all the different types of devices. CEM provides you with an interface to author flows and monitor them with ease. CEM is made up of a few components, namely Edge Flow Manager (EFM), and MiNifi. EFM provides you with a familiar user interface, similar to NiFi's, while MiNifi is used as the tool which helps you retrieve data from hard to reach places.
CEM also allows you to granularly deploy models to every different type of device in your enterprise
Prerequisites
- Deployed MiNiFi C++ agent on AWS EC2 Ubuntu 18.04 instance
- AWS: t2.micro or similar as a minimum
- Deployed CEM on a Cloudera DataFlow cluster
- Deployed a CDH Cluster with CDSW
Outline
Build Data Flow for MiNiFi via CEM UI
Publish Data Flow to MiNiFi Agent
Concepts
We will use Cloudera Edge Manager (CEM) to build a MiNiFi dataflow in the interactive UI and publish it to the MiNiFi agent running on the edge. This dataflow will ingest the car sensor data coming from ROS and push it to NiFi running in the cloud. In this tutorial you will simulate having our Smart Car by using a cloud VM--we instruct you to use AWS but feel free to use your favorite public cloud provider--which will serve as the MiNiFi agent.
To build out this pipeline we used three Virtual Machines in AWS
| Name | Service used | Size | OS |
|---|---|---|---|
| Edge-Smart Car | MiNiFi | t2.micro | Ubuntu 18.04 |
| Cloudera DataFlow | CEM | m5.2xlarge | Centos 7 |
| Cloudera Data Platform | HDFS + CDSW | m5.4xlarge | Centos 7 |
Build Data Flow for MiNiFi via CEM UI
To begin you will need the training data on the MiNiFi agent instance, ssh onto that instance as download the data
mkdir -p /tmp/csdv/data/input/racetrack/image/ && wget -O /tmp/csdv/data/input/racetrack/image/image.tar.gz https://github.com/hortonworks/data-tutorials/blob/dev/tutorials/cdf/edge2ai-autonomous-car/assets/data/image.tar.gz
tar -xvzf /tmp/csdv/data/input/racetrack/image/image.tar.gz
to change your MiNiFi configurations, if you are working with a new MiNiFi Agent copy these configuration by downloading them directly to the machine running MiNiFi
wget -O /home/ubuntu/nifi-minifi-cpp-0.6.0/conf/minifi.properties https://raw.githubusercontent.com/hortonworks/data-tutorials/dev/tutorials/cdf/edge2ai-autonomous-car/assets/services/minifi_cpp/minifi.properties
or by downloading them onto your local computer and sending them to the agent
wget -O ~/Downloads/minifi.properties https://raw.githubusercontent.com/hortonworks/data-tutorials/dev/tutorials/cdf/edge2ai-autonomous-car/assets/services/minifi_cpp/minifi.properties
scp -i /path/tp/pem ~/Downloads/minifi.properties <os-name>@<ec2-public-dns>:/home/ubuntu/nifi-minifi-cpp-0.6.0/conf
Edit the following properties of the minifi.properties file:
Table 0: Update minifi.properties
| Property | Value |
|---|---|
nifi.c2.agent.coap.host |
cem-public-DNS |
nifi.c2.flow.base.url |
http://cem-public-DNS:10080/efm/api |
nifi.c2.rest.url |
http://cem-public-DNS:10080/efm/api/c2-protocol/heartbeat |
nifi.c2.rest.url.ack |
http://cem-public-DNS:10080/efm/api/c2-protocol/acknowledge |
nifi.c2.agent.class |
AWS_agent |
nifi.c2.agent.identifier |
AWS_AGENT_001 |
Open your CEM UI at <cem-public-DNS:10080/efm>, if your minifi.properties configuration file is setup correctly you will find that your agent is sending heartbeats to the monitor events section of CEM UI
Now we know that our Agent can communicate with CEM, that is great news. In order to connect MiNiFi to NiFi we need to know where we are going so let's create a path for our data to slowly build a flow. Open NiFi UI on your CDF cluster and create a new input source named AWS_MiNiFi_CSV leave it alone for now, we will need the input id to set our connection from MiNiFi processor to NiFi Remote Process Group (RPG)
That is all we need to do on NiFi for now. Navigate to the Flow Designer on CEM UI, you can click on the class associated with MiNiFi agent you want to build the dataflow for, note that we named our agent AWS_AGENT_001 and our class AWS_AGENT
Note: Later when MiNiFi C++ agent deployed on separate the Jetson TX2, the class called "CSDV_agent" will appear.
For now click class AWS_agent. Press open to start building. The canvas opens for building flow for class AWS_agent:
We will build a MiNiFi ETL pipeline to ingest csv and image data.
Add MiNiFi Processors
Add a GetFile for CSV Data Ingest
Add a GetFile processor onto canvas to get csv data:
Update processor name to GetCSVFile.
Double click on GetFile to configure. Scroll to Properties, add the properties in Table 1 to update GetFile's properties.
Table 1: Update GetCSVFile Properties
| Property | Value |
|---|---|
Input Directory |
/tmp/csdv/data/input/racetrack/image |
Keep Source File |
false |
Batch Size |
1 |
Push CSV Data to Remote NiFi Instance
Add a Remote Process Group onto canvas to send csv data to NiFi remote instance:
Add URL NiFi is running on:
| Settings | Value |
|---|---|
URL |
http://<nifi-public-DNS>:8080/nifi/ |
Connect GetCSVFile to Remote Process Group, then add the NiFi destination input port ID you want to send the csv data:
| Settings | Value |
|---|---|
Destination Input Port ID |
<NiFi-input-port-ID> |
Note: you can find the input port ID by clicking on your input port in the NiFi flow. Make sure you connect to the input port that sends csv data to HDFS.
Add a GetFile for Image Data Ingest
Add a separate inport port on NiFi UI and name it AWS_AGENT_JPG just like before leave it blank for now. Now change to your CEM UI and add a GetFile processor onto canvas to get image data:
Update processor name to GetImageFiles.
Double click on GetFile to configure. Scroll to Properties, add the properties in Table 2 to update GetFile's properties.
Table 2: Update GetFile Properties
| Property | Value |
|---|---|
Input Directory |
/tmp/csdv/data/input/racetrack/image/logitech |
Keep Source File |
false |
Batch Size |
10 |
Push Image Data to Remote NiFi Instance
Table 3: Add a Remote Process Group onto canvas to send image data to NiFi remote instance:
| Settings | Value |
|---|---|
URL |
http://<cem-public-DNS>:8080/nifi/ |
Table 4: Connect GetImageFiles to Remote Process Group, then add the following configuration:
| Settings | Value |
|---|---|
Destination Input Port ID |
<NiFi-input-port-ID> |
Note: you can find the input port ID by clicking on your input port in the NiFi flow. Make sure you connect to the input port that sends image data to HDFS.
Publish Data Flow to MiNiFi Agent
Click on publish in actions dropdown:
Make this flow available to all agents associated with AWS_agent class, press publish:
Note: you can add comment
Sending driving log csv and image data to NiFi
Result of published successful:
Summary
Great, now you have completed the first step towards building a full data pipeline from a remote edge devive to a big data cluster. In the next section we will be connecting this flow to our cluster and set up our data for ML model training.