Tesla HW2 Autopilot Tips

I’m not sure I’ve ever enjoyed being stuck in freeway traffic before autopilot, but a few tips thanks to my adventures yesterday.

To engage the beta traffic-aware cruise control you need to be traveling at least 5 MPH. But once engaged, the system is quite happy bringing the car to a full stop and then starting up again. Perfect for California freeway stop-and-go traffic.

Low-speed autosteer also works great in freeway traffic under 35 MPH. However, if your cruise control is set above 35, which it typically would be on the freeway, it won’t work, even if you are slowed to below 35. So once you enter traffic, just click down the cruise to 35, then the autosteer icon should appear, then click one more time to engage.

I spent about 30 minutes yesterday in freeway stop-and-go traffic, 100% under autosteer + traffice-aware cruise control. While I won’t necessarily go seeking out traffic, the Tesla HW2 autopilot sure makes it more bearable. I can’t wait for future upgrades that allow speeds over 35.

Posted in AI Car, Tesla | Comments Off on Tesla HW2 Autopilot Tips

Tesla HW2 Autosteer First Drive

Yesterday evening I was driving on the CA 87 freeway in Santa Clara, heading South to San Jose with three other NVIDIA solution architects. It was after sunset, raining, and slow traffic even in the carpool lane. Then, alsmost as if on queue for Jensen’s CES keynote which we were off to watch, the low-speed autosteer icon appeared on my dashboard, indicating autosteer was available.

In current beta form, low-speed autosteer is only available when driving 35 miles per hour or slower. Traffic in the non-carpool lanes was stop-and-go, but the carpool lane was moving in the mid-30’s, perfect conditions to test out autosteer. Even at nighttime and in the rain, the system steered the car dead down the center of the lane. Driving manually, I would typically hug the left side of the carpool lane since that section of CA 87 has a wide left shoulder, so being a few inches closer to the stop-and-go traffic in the lane just to my right was at first a bit unnerving. After about a mile, however, I got used to letting autosteer do the work. The system is actually quite intuitive with just a little bit of practice.

After only a few miles I had to disable autosteer to merge across 3 lanes of traffic for my exit. As soon as I entered the slow lane, however, the autosteer available icon re-appeared so I gave it a try for the last half mile. Once again, the system worked flawlessly, even steering the car off the freeway and down the right lane of the three lane exit ramp at Park Ave.

The only thing more exciting than testing out autosteer was of course Jensen’s CES keynote where we heard about the latest innovations for AI for self driving cars, for your home, and in PC gaming. In case you missed the yesterday’s keynote, you can watch the replay below.

The NVIDIA channel on YouTube has the keynote nicely divided into segments, like the AI for Self Driving car segment.

Our you can watch the entire keynote, about 90 minutes, on UStream.


Posted in Uncategorized | Comments Off on Tesla HW2 Autosteer First Drive

Tejon Ranch to Gilroy With Tesla HW2 Autopilot

Headed North today on California’s interstate 5, with plenty of charge to make it to the Buttonwillow supercharger, I pulled into Tejon Ranch supercharger around lunchtime today not because of the nearby In-N-Out Burger but hoping I had driven enough for the cameras on my Tesla to have completed calibration. It can take from 1 to 5 hours of driving for the cameras to calibrate after you install the 2.50.185 firmware, you don’t get a message saying calibration is complete, and you have to stop and restart your car after calibration is complete for the new autopilot functionality to become available. But it was worth it, even without the burger. As I drove back onto CA 5 North the dashboard immediately lit up showing the availability of the traffic aware cruise control (beta) autopilot feature.

While the current low speed autosteer functionality only works below 35 MPH, the dashboard display clearly shows that the NVIDIA powered HW2 system is working to track the lane the car is driving down, preparing for future upgrades. The position of the car in lane is not a canned image, by all appearances it is being generated in real time as you drive down the lane. Steer towards the right or left of the lane and the display tracks your every move.

The beta autosteer feature, as noted, only works below 35 MPH and only on “restricted highways” (I have not figured out what that means yet), and I haven’t had a chance to try it out yet, so more on the traffic aware cruise control.

My last car had so called “adaptive cruise control”, which I used often, but simply can’t be compared to the Tesla’s conceptually similar feature. The Tesla system offers 7 settings which let you control how close you follow other vehicles by twisting the autopilot control stalk on the steering column. I found 3 was an ideal setting for the 70 MPH speed limit on CA 5. Later in the day, on CA 101 in the rain, I adjusted the setting to 4 to leave a little more space. The system worked flawlessly for all 225 miles from Tejon Ranch to Gilroy.

One of the biggest differences from my old car was perhaps not due entirely to the HW2 but to the instant torque of the Tesla’s electric motor. In my old car, when the system slowed down due to a slow moving car, it took ten or more seconds for the car to accelerate back to the cruise setting, something that no doubt raised the blood pressure of many a driver who grew impatient with my failure to instantly respond to a slow car ahead pulling over into the slow lane. On the Tesla, the response is near instantenous.

Of course in California, on long stretches of CA 5, you still have a few reckless drivers that will try to pass you even if there is barely a car length between you and the car ahead. This sends more commonplace adaptive cruise control systems into another brake and slowly re-accelerate cycle, but the Tesla system again responded flawlessly to the few rude drivers that tried to cut in front of me today.

Google Maps on my phone estimated 5 hours 30 minutes for my drive today, while the Tesla nav system, including two charging stops, showed 6 hours 10 minutes. Actual time was within 5 minutes. But I arrived at my destination significantly more refreshed thanks to having the opportunity to get out and walk a bit as my car recharged and thanks to the first autopilot features. I can’t wait to see what comes out in the next update. Keep up the great work, Tesla!

Posted in AI Car, Tesla | 1 Comment

Tesla HW2 Autopilot Upgrade (v2.50.185)

As Elon Musk promised in his New Year’s Eve tweet and exactly two weeks and 649 miles after taking delivery of my Tesla Model S, on New Year’s day I received my HW2 Autopilot software upgrade. The only nicer surprise was that the state of California managed to deliver my new license plates in even less time.

Tesla’s HW2 autopilot hardware has been installed on all new Tesla Model S and Model X vehicles manufactured after about mid-October 2016 and is enabled if you purchased the Enhanced Autopilot option. An accompanying Full Self-Driving Capability option should also be purchased if you want to take full advantage of the NVIDIA powered HW2 platform. Both options can be purchased at any time on vehicles manufactured after the mid-October cutover date if you did not configure at time of original purchase or lease. One warning, if you lease your vehicle and didn’t select the option, you will need to pay the full cash price of the option[s] to upgrade. You can’t just add the option onto the lease payment later.

The CES show this year promises to be full of companies demonstrating self-driving capabilities, many of them powered by NVIDIA. You can preview NVIDIA happenings at the CES show highlighted in the teaser video below.

But back to the HW2 autopilot upgrade. If you want self-driving capabilities in a production vehicle you can actually purchase today, Tesla is your only option. First of all, while the press have rumored this upgrade to be called 8.1, Tesla still is calling it 8.0 with the accompanying firmware version of 2.50.185. I also expect the install warning of “will take 1 hour 40 minutes to install” is a carry-over from the original 8.0 upgrade as this install took only about 20 minutes to complete. It is important to note that unlike traditional vehicles which remain the same as the day you drove off the lot, Tesla is able to continuously upgrade their vehicles through software upgrades delivered wirelessly to your car. This allows for incremental updates so you can expect more and smarter features to be added over time. So what’s in this first HW2 upgrade?

Forward Collison Warning

The first new feature to be enabled is forward collision warning. While collision warning systems are available in quite a few new vehicles today, the HW2 powered system promises to be more accurate, with fewer false alarms, than traditional systems. And unlike the system that may be hardwired into your car today, the Tesla system can be upgraded over time.

Traffic-Aware Cruise Control (Beta)

Tesla chose to add the “Beta” label to this feature, indicating it isn’t fully complete. Again, conceptually this feature is similar to existing radar-based adaptive cruise control systems but promises additional capabilities based on the ability of the HW2 software to understand more about the complete environment on the road around you.

Low-Speed Autosteer (Beta)

The third major feature enabled by the upgrade also gets the Beta label. Working in conjunction with traffic-aware cruise control, this software helps you maintain your position in lane. Here is where the power of the Tesla HW2 approach really starts to become apparent. Each feature is simply a software application running on the NVIDIA HW2 self-driving car supercomputer buried deep inside the car. So having one feature or program communicate with another is as simple as sharing information between two applications on your laptop or smart phone

Unfortunately, you will have to wait until my next blog update to hear my driving impressions of these new autopilot features. I only drove a few miles yesterday after completing the install, and as I pulled into the driveware the system was still reporting “calibrating cameras – autopilot not yet available for use”. Meanwhile, here are a few other articles on the upgrade.

Complete Release Notes (courtesy Electrek)

Inverse article

Teslarati article with more screen shots

Tesla Motors Club article – cameras calibrated after about 100 miles

Reddit thread

Update 1

Drove about 90 minutes/100 miles today, amazing how good the LA freeways are on holidays. Cameras still calibrating. I sent email to Tesla’s autopilot feedback asking how long the camera calibration takes and received a prompt and helpful reply.

“The process for the calibration takes about one to five hours of driving. Calibration is quicker on well-marked roads, but road conditions (not marked, poor visibility, etc.) may significantly increase the time required. The features will not be enabled until the next drive cycle on the vehicle once calibration is complete.”

Posted in AI Car, Tesla | Comments Off on Tesla HW2 Autopilot Upgrade (v2.50.185)

Tesla v8.0 (2.50.180) Update

On Christmas morning, I awoke to find a minor software update waiting for my 8 day old Tesla Model S. While the install screen warned me the update would take 1 hour 40 minutes, it actually completed much quicker. Being Chrismas morning I was otherwise preoccupied and didn’t time the update, but glancing at my mobile phone about an hour later I noticed the update had already completed. My speculation is that the 1 hour 40 minute time is for older models that don’t have the new and 40x more powerful NVIDIA powered AutoPilot 2.0 (AP 2.0) hardware.

This is not the much awaited 8.1 software update which will enable the AP 2.0 autopilot capabilities, but it does seem to add some test functionality to the car’s new NVIDIA hardware. More on that later.

First the official stuff.

The “Automatic Driver Profiles” is a nice touch that links the key fob to a driver profile. Link your key to your profile, give the #2 driver the same treat, and make everyone else have to select their driver profile on the main screen. Now saving 1 click when entering the car doesn’t seem like a big deal, but since many cars have a similar feature, it is a nice to have. The main difference here is that Tesla enabled the feature with a simple software upgrade. Try doing that on your favorite luxury car…

Rounding out the minor update are inclusion of supercharger amenity icons and additional equalizer bands, the latter lost on my tone-deaf ears. But my kids may use that feature when they borrow the car.

Now back to the AP 2.0 hardware. The 2.50.180 release notes don’t mention anything about it, but the dashboard display clearly does. When driving next to other cars, sensor icons on each corner of the car now light up, white, yellow, then red to indicate nearby objects. This added dashboard display feature is informational only at this time, but clearly will be feeding into the autopilot software when that feature is enabled. A bit more on this can be found in the Tesla forums.

Now for the fun stuff. I’m not sure if this was part of the 2.50.180 upgrade or not, but as reported elsewhere, pressing and holding the Tesla icon at the top of the main screen and typing in the code “mars” places the navigation system somewhere on the red planet, complete with the expected pan and zoom capability.

Returning from the Mars navigation screen, I noticed my Tesla was also temporarily upgraded to a different model.

Posted in AI Car, Tesla | Tagged | Comments Off on Tesla v8.0 (2.50.180) Update

The Perfect GPU For The Holidays

If your shopping for that perfect holiday gift, there is still plenty of time to buy a new NVIDIA Titan X or other Pascal based GPU. This year however, I decided on an NVIDIA GPU in a slightly different package, which found its way to my doorstep yesterday.

In October, when Tesla announced all new cars being produced would have full self-driving hardware, I first started to look at the Tesla Model S for the soon to be expiring lease on my current car. According to Tesla CEO Elon Musk, the car’s “Tesla Neural Network”> is powered by an NVIDIA GPU which gave me one more reason to consider the car.

Many years ago, when I bought my first car, the main subject of conversation among my friends was typically about how much horsepower it had. With Tesla now shipping the first real AI car, new car buyers would be better served by asking how many TOPS (trillions of deep-learning operations per second) the car comes with. While Tesla doesn’t specify the exact TOPS spec for the car, they do say the new NVIDIA gear is 40 times more powerful than the previous autopilot hardware.

Today I still had to drive my new Tesla the old fashion way, with hands and feet, as the new autopilot software isn’t quite ready. Shortly, a new OTA download should be available enabling the autopilot. As you might expect, I of course got both the “enhanced autopilot” option as well as the “full self driving capability” option, which promises plenty of new self-driving features over the coming months.

I already promised my team of solution architects working on the NVIDIA DriveWorks AI car software a chance to test out the Tesla updates as soon as they are available. During the winter break, I also plan to test out the Tesla Supercharger network with drives from Santa Clara to Los Angeles and then to Portland. I’ve already given away 1 of my 10 referral discounts, but if anyone out there is thinking of buying a new Tesla this month and wants a $1000 discount, I have 9 left. Take note, however, the free supercharger network won’t be totally free anymore for vehicles purchased after January 1, 2017, as Tesla announced last month.

Holiday errands are going to be a lot more fun this year!

Posted in AI Car, Tesla | Comments Off on The Perfect GPU For The Holidays

GTC India Keynote

NVIDIA’s Shanker Trivedi kicked off the opening keynote at GTC India today.

One of Shanker’s key messages to India’s GPU developers was to make sure to sign up for our Developer Program to get access to all the latest NVIDIA SDKs, CUDA 8 software, and other great developer tools. I’m surprised how often I talk to developers who aren’t registered and using versions of NVIDIA software that are months or years out of date.

Speaking of developers, as promised, one of my favorite apps from yesterday’s Emerging Companies Summit is Recipe Book. Apparently at least a few developers like to cook, and I remember online recipe sites from the early days of the Web. Recipe Book is a lot more than just recipes. It uses GPU deep learning to suggest recipes you can make given a list of ingredients on hand. It can even suggest recipes based on a picture of what you have in your refrigerator. I can’t wait to get home and try this app out. Given my travel schedule, I often find myself trying to figure out what to make the day before a business trip when I don’t want to go to the store and restock.

Posted in Uncategorized | Comments Off on GTC India Keynote

GTC India 2016

NVIDIA’s GTC India 2016 starts Tuesday in Mumbai but today is busy with preshow events including a packed set of NVIDIA DLI (Deep Learning Institute) sessions at IIT Bombay and this afternoon’s Emerging Companies Summit. Some of the hottest AI and Deep Learning startups will be presenting at ECS and I’ll update this post later with some of my favorites.

While I flew into Mumbai, I expect at least a few GTC attendees will arrive via Mumbai’s iconic Victoria Station, a Unesco World Heritage Site.

For more photos from GTC India, visit my SmugMug site.

Posted in Uncategorized | Comments Off on GTC India 2016

Energy Efficient Supercomputing with DGX SaturnV

Just how energy efficient is NVIDIA’s new DGX SaturnV? At about 9.5 gigaflops/watt, the Pascal P100 powered system is about 50% more energy efficient than the nearest non-NVIDIA powered system. But unless you follow the Top500 list of world’s fastest supercomputers, it probably is hard for you to visualize if 9.5 gigaflop/watt is a good or not. But many have taken note of NVIDIA’s 50% jump over competitors in efficiency. Think of buying a Tesla Model X that can go 300 miles on a single charge and then hearing a competitor had released a car capable of 450 miles. You would take note.

But back to DGX SaturnV’s intended use supporting NVIDIA’s AI software development. Does energy efficiency matter there? At a recent GTC, Baidu’s Andrew Ng stated that it cost about $100 in electricity alone to train their Deep Speech 2 Deep Neural Network (DNN) for English and Mandarin speech recognition. It is not uncommon for a data scientist to train a DNN 50 or 100 times before getting good results and Baidu has 100’s if not 1000’s of data scientists. The energy savings of a Pascal P100 powered system can add up quickly. All of us involved in the rapidly growing field of AI need to take energy efficiency into account when thinking about how to power our research.

DGX SaturnV is built out of 124 DGX-1 deep learning supercomputers each in turn with 8 Pascal P100 GPUs. But you don’t need to build the world’s 28th fastest supercomputer to take advantage of the Pascal P100’s energy efficiency. Last week, NVIDIA and Microsoft announced record breaking results for Microsoft’s Cognitive Toolkit running on DGX-1, an amazing 170 times faster than a standard two-socket x86 server. And if your budget can’t afford a DGX-1, you can get started with Microsoft’s Cognitive Toolkit using Azure’s new N-series GPU virtual machines for as little as 56 cents/hour. And of course in-between you have other options including the Titan X Pascal, the most powerful GPU available for PC developers.

Posted in Uncategorized | Comments Off on Energy Efficient Supercomputing with DGX SaturnV

Tesla Shows Off The AI Car

The AI Car is here. Tesla’s latest video shows the capabilities coming soon to every new car now shipping from the Tesla factory.

Electrek has a more complete analysis of the video. This earlier article by Electrek detailed the NVIDIA GPU technology inside Tesla’s latest cars.

While all new Tesla cars rolling off the assembly line already include the NVIDIA hardware to power the capabilities demonstrated in the video, Tesla will continue to roll out new over the air software updates over the coming months, adding new autonomous capabilities with each release.

It will be interesting to see how fast other automobile manufacturers can introduce similar capabilities.

Posted in AI Car | Comments Off on Tesla Shows Off The AI Car