Bus tracker hardware

[luqmaan]

What problem are we trying to solve?

CapMetro publishes vehicle positions every two minutes. This is better than nothing, but in an ideal world we should have updates every five seconds. Because CapMetro uses the LCRA radio system to publish vehicle positions, ~2 minutes is the max frequency for every vehicle to report its position.

Lets try building some hardware to monitor the bus positions ourselves. This doesn't seem completely impossible:

• Raspbery Pi, Arduino, etc. are cheap.
• Software like transitime can automatically assign vehicles to trips. This means that the driver of a bus doesn't need to tell the hardware tracker that it is currently on Route 801.
• Data isn't that expensive. https://ting.com/ costs ~$10/mo for 500MB of data.

The ultimate goal of this project idea isn't to build something production ready. It is to build a prototype/proof of concept and see what opportunities arise.

Who will benefit (directly and indirectly) from this project?

People who use the bus and need to know where the bus is.

Links to any research/data available/ articles

Code for Miami has started already: https://github.com/herrdragon/busTrackingGps

What are the next steps (validation, research, coding, design)?

Research.

What help is needed at this time?

• Who else is working on building hardware like this? Are there any companies doing this?
• Are there any groups in Austin that do things with hardware interested in this idea? (Maybe Engineers without Borders)

[scascketta]

Ignore this, see the comment below:

I started working on this idea this week. I've got an old Raspberry Pi Model B+, a USB GPS receiver, and a USB 3G modem. I'm working on a Go program to use a publish-subscribe protocol called MQTT to send updates to AWS IoT (which has a managed MQTT broker) every 5 seconds. AWS IoT can then forward the updates to DynamoDB using its rule system, and we can write a Lambda function to consume the updates from DynamoDB and produce a GTFS-realtime feed containing vehicle locations.

Communicating with the GPS receiver is quite easy using gpsd, a daemon that provides, among other things, a JSON interface for the GPS data on a local Unix socket.

Powering the Raspberry Pi is pretty straightforward, any USB battery that provides enough amperage for the board and the accessories would be sufficient. I think 2 amps is the minimum necessary. Someone has measured power consumption for a variety of Raspberry Pi models.

Establishing connections with the 3G data modem is a bit trickier. I've been able to use Sakis3g to connect to the Internet over Ting's network but I've had issues getting it to persist connections in the event of signal loss or system reboots. I could really use help with this part, if anyone has experience with this. With Ting's rates, using 500MB/month would cost $18/mo.

In the future, I also want to make reproducible Raspbian images with the GPS update daemon and 3G scripts baked into the image. To start with, it's easy enough to just make an image of the Pi's SD card, and then distribute that, but ideally we could customize the Raspbian unattended net installer.

[scascketta]

...or we could use a cheap Android phone and just write an Android app to speak to AWS IoT, and let Android do all the hard work with cellular data connections and GPS. Chime in if you have Android experience.

The Android app can use MQTT to send updates to AWS IoT (which has a managed MQTT broker) every 5 seconds. AWS IoT can then forward the updates to DynamoDB using its rule system, and we can write a Lambda function to consume the updates from DynamoDB and produce a GTFS-realtime feed containing vehicle locations.

[scascketta]

I built a scrappier version (repo) of what I posted above a couple weeks ago. The location tracking works fine (simple to begin with). Instead of using MQTT to contact AWS IoT, it's now just HTTP -> API Gateway -> Lambda -> DynamoDB. The next steps are:

• [ ] Minimize data usage (currently at around ~80MB/day).
  • I think requesting location updates exclusively from the GPS receiver will save data (at the cost of some battery, and a longer time-to-acquire-location).
  • Could also try to use MQTT again, but it's a real pain to get working.
• [ ] Identify the vehicle's current GTFS trip automatically so the operator doesn't have to update it manually.

Oh also, with some input from @luqmaan, I emailed CapMetro for their feedback on this idea back on April 2nd, 2016:

I've been thinking about how to report vehicle positions more frequently, for example, every five seconds. I've set up some proof of concept hardware: https://github.com/open-austin/project-ideas/issues/63. But before I work on this more, I'd like to know what you think of the idea. What are the bureaucratic barriers? What are the technical barriers? Do buses have anything like a AC/DC outlet or USB port onboard? I'd really appreciate your feedback on this idea.

I pinged them again today and got a response:

Capital Metro is not interested in pursuing additional CAD/AVL hardware that would not integrate with our existing systems at this time. We are working to improve the polling rate of the vehicles in the coming fiscal year and hope to continue to improve our service and the availability of our data. Unfortunately, we cannot allow any third party devices to be installed on our vehicles.

[3vivekb]

At the @vta the CAD/AVL system produces gps info 90 seconds slow (over 1080 baud modems) but they pump their numbers into some proprietary Trapeze / Xerox software that made rather decent predictions. Not every two minutes mind you.

Instead though the buses got a Icomera/Moovit wifi routers installed that gives instant gps and speed data. And we're gonna build our real-time off of that.

TLDR: maybe it's not a question of improving their CAD/AVL but instead tapping into their wifi provider?

[luqmaan]

Trapeze / Xerox

Thats what CapMetro uses as well.

tapping into their wifi provider?

Most of the routes don't have WiFi unfortunately.

[mscarey]

Closing this because I think y'all succeeded in proving your point with this project.