Tesla Smart Summon patent highlights progress in 3D labeling for full self-driving features

Tesla Smart Summon in action. (Credit: Rody Davis/YouTube)

A recently published Tesla patent application details the machine learning methods behind Smart Summon, specifically highlighting the progress being made with 3D labeling in training data.

The application, titled “Autonomous and User Controlled Vehicle Summon to a Target,” utilizes machine learning methods explicitly detailed in two other recent Tesla patent publications in its functionality. This series of three inventions altogether describes an automated way of generating training data which is then used by a machine learning model to accomplish an expansive list of self-driving capabilities in Summon.

“Traditionally, much of the effort to curate a training data set is done manually by reviewing potential training data and properly labeling the features associated with the data,” Tesla’s first application in the series states. “The effort required to create a training set with accurate labels can be significant and is often tedious… Therefore, there exists a need to improve the process for generating training data with accurate labeled features.”

A method flow chart from Tesla's autonomous 3D labeling patent. | Image: Tesla/USPTO
A method flow chart from Tesla's Smart Summon patent application. | Image: Tesla/USPTO

The application goes on to describe how labeled training data is made autonomously in their invention using sensors and the collection of what’s called a “time series,” i.e., a series of images captured over a period of time.

“Using data captured by sensors on a vehicle to capture the environment of the vehicle and vehicle operating parameters, a training data set is created,” it explains. “In some embodiments, a three-dimensional representation of a feature, such as a lane line, is created from the group of time series elements that corresponds to the ground truth… As one example, a series of images for a time period, such as 30 seconds, is used to determine the actual path of a vehicle lane line over the time period the vehicle travels…a single image of the group and the actual path taken can be used as training data to predict the path of the vehicle.”

Tesla CEO Elon Musk has previously mentioned that better labeling is one of the keys to speeding up the rollout of self-driving functionality and features like Reverse Summon. “We need to finish work on Autopilot core foundation code & 3D labeling, then functionality will happen quickly. Not long now,” Musk wrote on Twitter in March this year. With better labeling (more accurate training data) comes safer and more capable software due to improved predictions from the modeling.

Tesla Owners Silicon Valley Smart Summon Model 3s (Credit: @MinimalDuck)

When it comes to Tesla’s Smart Summon, prediction modeling is essential considering there isn’t a driver in the vehicle during its operation. The patent publication covering Summon embodies the first application’s time series functionality and a second application’s implementation of the time series’ training data in its methods, demonstrating one of the numerous potential uses for the machine learning invention. Hints about future developments using Smart Summon are also detailed in the application. Examples include:

  • Syncing the Smart Summon with a calendar so the vehicle “automatically navigates to arrive at the location at the ending time, such as the end of a dinner party, a wedding, a restaurant reservation, etc.”
  • Implementing a multi-part destination into the Summon instructions such as waypoints at an airport to pick up multiple passengers.
  • Monitoring the heartbeat of a Summon user to ensure they are maintaining a connection with the vehicle while operating the feature.
  • Customizing the vehicle’s arrival settings such as interior lighting, exterior lighting, hazard lights, welcome music, and climate control preferences.

One of the more unique bits about the Smart Summon patent application is the appearance of Elon Musk as an inventor. While the CEO is known to be intimately involved in nearly all aspects of vehicle design, software features, and business operations, his name is unexpectedly absent from most of the company’s inventions. However, this is apparently on purpose. “I generally try my best not to be on patents,” he revealed on Twitter in reply to a post about the Smart Summon application. Notably, inventorship is a legal definition based on the conception of an invention, i.e., not the person/people who suggested or directed its creation, but the person/people who devised the means to accomplish it.

Prior to the most recent patent publication, Musk contributed inventorship to the door and body styling of the Model X. He also contributed the same to both the design and function of Tesla’s vehicle charge inlets.

Dacia J. Ferris: Accidental computer geek, fascinated by most history and the multiplanetary future on its way. Quite keen on the democratization of space. | It's pronounced day-sha, but I answer to almost any variation thereof.
Disqus Comments Loading...