Thanksgiving Break

Looking forward to a hopefully relaxing week, spending time with my family, and working on pushing some of my projects forward. Here are some of the things I have on my list.

Road Surface Detection
I have been working on a real-time AI/ML solution for predicting coefficient of friction based on the input from an optical camera. I am using an NVIDIA Jetson Xavier AGX (thank you ADLinkTech) , I have trained a SSD_mobilenet, and but would like to see if I can get some better performance (lower latency), and improved accuracy, so I am porting it over to the Google Cloud AutoML Vision Edge, this will allow me to test a bunch of different ML models automatically without having to run it through my ML pipeline based on Kubeflow, I am excited to see the results.

CAN bus CDDS, and FogLamp
Part of the solution I have been developing uses CAN bus data to do validation, and verification. I’ve been using FogLamp to help manage the data streams, as part of this solution, I am going to continue work with Eclipse Cyclone DDS, and the ROS2 implementation, the plan is to get all of the sensor data from CAN bus to CDDS, then use FogLamp to send everything northbound.

Cyclone DDS is a standard-based
technology for ubiquitous, interoperable,
secure, platform independent, and real-
time dat...

Simulation Environments
A lot of my testing happens in simulation, I started out with IPG which I am very happy with in terms of performance, and accuracy, however, I like to keep my options open, and have been working on porting my base model to the Simulink Vehicle Dynamics Blockset, while I was at it I decide to survey the landscape a bit more, and see what other platforms are available, it looks like some of the game engines are starting to provide robust simulation environments, so I am also testing Unity3d Simulation.

Radar and ground truth

Spec MX5
Going to start working on the build sheet for my spec MX5, found a really nice donor down in San Antonio. I am going to drive her as a street car for a couple more weeks, hoping to get out to Harris Hill Raceway once or twice, to work on my heel-toe.

Decisions for 2020 – Back to Basics

The plan for 2020 is to get a lot more track time. I need to work on my technique. Over the last two years with my GTB1, I have only been driving practice sessions, and races. I am looking to see if I can find the holy grail, a completely race ready car that is streetable, probably too much to ask. I want to be able to drive down to Harris Hill, COTA, Driveway, MSR Houston, or Cresson do DE’s or work with an instructor, then drop off the car to do an arrive/drive for long distance races. I want something that has a clear class definition with all of the sanctioning bodies (SCCA, NASA, WRL, AER). Additionally, I am interested in learning the racecraft associated with running in a Spec class. The GTB1 class I have been running with in PCA is a modified class, and can feel a bit like an arms race.

So far my research has led me to two possible choices, which have similar price points, and are philosophically aligned to my goals: Spec MX-5, and Spec E46. Both have reasonable price points both in terms of the up-front cost, and TCO. Both are well defined, with lots of cars on the market. So part of the decision will be build vs buy. I am leaning toward my previous approach which was to buy something proven.

Along side the desire to go Spec racing, and have a streetable track car, I am also going to pursue renting a seat in a couple different classes within the SRO, or possibly even the upcoming Creventic 24hour race here in Austin. There seem to be a number of seats available in TC, TCR, or GT4 cars. I am interested in learning how to drive a touring car, there might be a good opportunity for this, more to come on this topic at a later date.

The GT4 program that Danny, and I were planning is not going to happen, we could not close the funding gap, which was significant. Here are a few stats on what we were able to accomplish between Jun, and Oct of this year. We prospected over 125 opportunities, of that number we disqualified 35 (with the goal of: “get to no, fast”). We closed contracts with 10 partners, and have an additional 3 contracts pending, we progressed 17 deals into contract/negotiation, and built a number of very strong partnerships along the way. These partnership will bear fruit over the course of 2020.

Our goal was a 1-2% conversation rate on 1000 opportunities, at just slightly over 10% of the total opportunities prospected our conversion rate was 12.5%, not too bad. The real challenge was prioritization. I have a full-time job, a start-up, a family, racing, and my health. Adding the sponsorship/partnership activity on top of all that was very challenging from a time management perspective. I am happy with what we accomplished, and learned a lot about the business side of motorsports. I plan to continue down a more focused path for 2020.

Practice, Product, and Progress

I’ve been spending time in iRacing learning the line, got both of David Murry’s excellent Sebring, and Car Suspension webinars, as well as reviewing playbacks from the VRS datapacks. I really look forward to working with David directly on the Test/Tune day before the weekend at Sebring. I was also able to spend some very productive time this week out at Harris Hill with Andris of Gradient Racing. More seat time, and getting everybody familiar with the car is always a good thing

David Murry in the BGB #38

Over the holiday, I got a lot of work done on the trackNotes app, it is nearly ready for an alpha release. I am using creator, for the front-end, and firebase for the backend. I put together some wireframes for another app I am developing called setupBoard, that will work either standalone or together with trackNotes. Regarding my work with Kubeflow, it has gone very well, I was able to train against my dataset with multiple models, distributed across three cloud providers: GCP, AWS, and IBM. Inception v3 is the current leader with predictions of 85% accuracy against the ground truth in testing. I am going to test a couple more models: Random Forest, a TF Boosted Tree, and a DNN Classifier. Once I have another round of training, and testing I will start working through AI assisted hyper-parameter tuning to see if I can get the error rate any better. At that point I will switch gears and start working on the front-end.

setupBoard wireframe in creator.ionic.io

On the health front, I was having a hard time finding someone who could help me with both my nutrition, and training. I was referred to Phil Gephart who immediately exceeded my expectations for his knowledge, and professionalism. Phil is awesome! He worked with me to switch around my macros, added some supplements, changed up my training program, I am already seeing some incredible results with a 3% drop in body fat within the first week!

Over the Holidays

I have always loved this time of year, it has given me a chance to focus on the things that I really want to do, interesting projects, time to catch up with friends, and family, play games (or get in some iRacing time, got to get Sebring down), watch movies, and just relax. I am off from work for the next couple weeks, and will be focused on a quite a few things, as usual.

I have built, and deployed three kubernetes clusters on the three different cloud providers that came with my having joined FFL (GCP, AWS, and IBM), I have tested all of my training jobs, and am now ready to do a big run on distributed training using kubeflow, I will be working on a couple different models, Inception v3 to start, and will be working through all the hyperparameter tuning involved with trying to get the best RMSE for what I am doing, hint (it’s motorsports related).

Also, I have a prototype of the trackNotes app, that I released to the Play Store for a closed Alpha, to get some feedback on the UI, I am putting it together with creator.  I am working on the backend data-model in Firebase, which is pretty straightforward, there are some really good materials out there, I really like the codelab, it is ridiculously comprehensive. I am also going to start a wireframe for another app called setupBoard.

On the racing front, I am doing outreach for sponsors, and will need to get all of that loaded into Hubspot so I can keep track of it all. I am also loading in a bunch of leads which, I will see if I can turn into opportunities for everything that I am doing (vaguely described above). Also, I have been talking to Andris over at Gradient Racing to see what support would look like for 2019, we are planning a test day in January.


Preparations for 2019

Continuing my work on track notes, I have a working prototype, just need to clean up the UI, and then start working on the iPhone cross-compile, I hope cordova works well for that.

I started a new 6-week workout block with two HIIT days, it is pretty rough from a recovery perspective, I do 45-minutes of HIIT cycling on Monday/Wednesday with 30-minutes of weight training, on Tuesday/Thursday I run, and do weights for 30-minutes each, on Friday I swim, and then do endurance focused weight training. Talking to my nutritionist about a Ketogenic plan to go with my training platform.

I keep adding to the backlog of reading materials, going through the bibliography of Michael Krumm’s excellent:  Driving On The Edge: The Art and Science of Race Driving

I have made some significant breakthroughs on my upmarket product, currently building out the underlying development platform, and pipeline using the cloud credits from FFL.

I am dying to get out in my car for a test day, looking at the next edge addicts event, or maybe just a session out at Harris Hill (which would be my preference). Hopefully, I can get that setup over the next couple weeks.