Tag: selfdriving

Selfdriving risk budgets

Their plan involves a risk-based analysis, aiming for “better than human” but not perfection. With this approach, you measure the acceptable risk, and you calculate the uncertainties in the various components of your system, such as perception, motion planning, mapping, and others. Each system has some extra capability to make up for uncertainties at other levels.

Tesla selfdrving

In shadow testing, a car is being driven by a human or a human with autopilot. A new revision of the autopilot software is also present on the vehicle, receiving data from the sensors but not taking control of the car in any way. Rather, it makes decisions about how to drive based on the sensors, and those decisions can be compared to the decisions of a human driver or the older version of the autopilot. If there is a decision — the new software decides to zig where the old one zags, or the new software cruises on when the human hits the brakes, an attempt can be made to figure out how different the decisions were, and how important that difference is. Some portion of those incidents can be given to human beings to examine and learn if the new software is making a mistake. If there is a mistake, it can be marked to be fixed, and the testing continues.

User Input is an error:

Elon Musk views any human user intervention is an error situation for the Tesla Autopilot. Elon means that whenever a human has to take control from the Tesla Autopilot system this is indicating an error that must be fixed for a future fully autonomous car.

Teslas improve with use:

Most of the systems we currently use aren’t built to improve through use. They have locked in performance and capabilities. These systems can only improve through revisions and patches made by technical experts. That approach is on the way out. Systems can now be improved operationally …. Further, for the most complex activities, this will be the only type of system you will be able to buy.

Let me guess, the media won’t be falling over themselves to report on these instances where the tesla autopilot saved lives.

Doctors told Neally later that he’d suffered a pulmonary embolism. They told him he was lucky to have survived. If you ask Neally, however, he’ll tell you he was lucky to be driving a Tesla. As he writhed in the driver’s seat, the vehicle’s software negotiated 30 highway km to a hospital just off an exit ramp. He manually steered it into the parking lot and checked himself into the emergency room, where he was promptly treated. By night’s end he had recovered enough to go home.

Another analysis on the Tesla software disruption:

Tesla’s first bet is that it will solve the vision-only problem before the other sensors get small and cheap, and that it will solve all the rest of the autonomy problems by then as well. This is strongly counter-consensus. It hopes to do it the harder way before anyone else does it the easier way. That is, it’s entirely possible that Waymo, or someone else, gets autonomy to work in 202x with a $1000 or $2000 LIDAR and vision sensor suite and Tesla still doesn’t have it working with vision alone.

The second bet is that Tesla will be able to get autonomy working with enough of a lead to benefit from a strong winner takes all effect – ‘more cars means more data means better autonomy means more cars’. After all, even if Tesla did get the vision-only approach working, it doesn’t necessarily follow that no-one else would. Hence, the bet is that autonomous capability will not be a commodity.

This video from 2014 is what happens when you improve cars at the speed of the software industry. very very impressive.

Being able to update the fleet isn’t just useful for selfdriving

Researchers Hacked a Model S, But Tesla’s Already Released a Patch If you were CEO of a car manufacturer, which of these headlines would you rather were written about you? The first speaks of a tired, old manufacturing model where fixes take months and involve expense and inconvenience. The second speaks of a nimble model more reminiscent of a smartphone than a car

Visual Object Detection

2 Cheap Cameras Can Provide LiDAR-like Object Detection

For most self-driving cars, the data captured by cameras or sensors is analyzed using convolutional neural networks – a kind of machine learning that identifies images by applying filters that recognize patterns associated with them. These convolutional neural networks have been shown to be very good at identifying objects in standard color photographs, but they can distort the 3D information if it’s represented from the front. So when Wang and colleagues switched the representation from a frontal perspective to a point cloud observed from a bird’s-eye view, the accuracy more than tripled.

Waymo Phoenix

Waymo has been testing self-driving cars without any safety drivers in Phoenix for nearly a year. Waymo will launch the first commercial program in Phoenix within months. They have applied California regulators to begin testing without safety drivers. The commercial self-driving in California will likely start in 2019.

Steps to self-driving

This is why so much work is going into how the vehicle might communicate with the user – ‘this is an L5 journey and you can sleep’, or ‘I’ll drive myself for the next hour, and alert you 5 minutes before it’s time for you to take over’? Does that autonomous golf cart just refuse to cross an invisible line into a neighborhood where it’s not certified for autonomy? And can you push the Johnnycab driver out of the way?

Waymo World Simulation

At any time, there are now 25000 virtual self-driving cars making their way through fully modeled versions of Austin, Mountain View, and Phoenix, as well as test-track scenarios. Waymo might simulate driving down a particularly tricky road 100Ks of times in 1 day. Collectively, they now drive 13M km per day in the virtual world. In 2016, they logged 4B virtual km versus a little over 5M km by Google’s IRL self-driving cars that run on public roads. And crucially, the virtual km focus on what Waymo people invariably call “interesting” km in which they might learn something new. These are not boring highway commuter km.

Self-driving disruption

By 2030, 95% of US car km traveled will be in self‐driving, electric, shared vehicles. Shared self driving cars will be 4-10x cheaper / km than buying a new car, and 2-4x cheaper than operating an existing paid-off vehicle. Global oil demand will peak at 100M barrels / day by 2020, dropping to 70M barrels / day by 2030 (same as 1995). Productivity gains as a result of reclaimed driving hours will boost GDP by an additional $1 trillion. the number of passenger vehicles on American roads will drop from 247M to 44M, opening up vast tracts of land for other, more productive uses.

Robot delivery

Amazon, which currently charges a $99 annual fee for 2-day deliveries under its “Prime” service, will eventually offer 2-tier pricing for delivery services. One will be a “Gold Prime” membership costing $199 to $249 a year that covers next-day deliveries, the other a platinum membership for $399 a year that includes same-day deliveries.

2019-08-20: Starship Technologies

The company has made over 100K commercial deliveries. The total funding has reached $85M. Parcels, groceries and food are directly delivered from stores, at the time that the customer requests via a mobile app. Once ordered the robots’ entire journey and location can be monitored on a smartphone. Starship delivery bots use machine learning to detect objects and do not use expensive LIDAR. Starship robots mostly drive on sidewalks and cross streets when they need to. This poses a different set of challenges compared to self-driving cars. Traffic on car roads is more structured and predictable.

2021-01-27: 1m now:

Starship reports that while its operation has not been flawless and its robots are always learning, any potential issues with the robot have not resulted in any injuries due to the low speed on the sidewalk. In addition to sidewalks, the robots are also doing 50K street crossings per day.

This might seem mundane, but both sidewalks and bike lanes are a huge opportunity. Even Amazon realized this, and is using both for last km delivery.