The Japanese brand wants to write the ultimate self-driving algorithm.
Honda has filed a patent for a system that will help future autonomous vehicles forecast the actions of other road users more accurately. CarBuzz discovered the patent, filed with the United States Patent and Trademark Office.
The original patent dates back to November 2019, but it has been updated twice as Honda's autonomous systems become more advanced and readily available in production cars.
To accurately predict what other road users will do in any given scenario, Honda must first build up a giant database of information that ego vehicles provide. An ego vehicle is equipped with sensors, LiDAR, cameras, or a combination of all three.
This hardware is then used to create a perceived environment to which the car responds. A basic example is the sensors attached to the front of the Honda Accord, which are responsible for activating forward collision avoidance and controlling the intelligent cruise control.
Honda states that most forecasting algorithms rely solely on stationary cameras, drones, social norms, scene context (like a pedestrian crossing), and semantical studies. According to Honda, more is needed. To build an actual self-driving system, a car has to have the same situational awareness as a human.
To achieve human-level thinking, Honda wants access to the image and dynamic data an ego vehicle collects. To put it another way, Honda wants recorded footage from your car and wants to see how other people and cars act around it. Over a predetermined period, the ego vehicle will collect vast amounts of dynamic data and footage. As you might know, the kind of data we're talking about requires a lot of space, so Honda proposes an onboard storage unit separate from all existing onboard storage units like the ECU. A self-driving car equipped with cameras can easily collect a terabyte of data per day.
These vast amounts of data can then be used to create the ultimate human-level algorithm, creating a neural network for all Honda vehicles with access to it.
We don't see this reaching production vehicles soon, simply because Honda needs time to gather data from existing cars before it can create the so-called neural network, which consists of a neural network processing unit and a localization dataset. The neural network communicates with the proposed future vehicle, arming it with human-level thinking and forecasting abilities.
As you can see in the images, the ego vehicle will be able to understand various scenarios that are possibly not understood by algorithms designed using only stationary cameras and social norms.
Honda classifies external objects into several categories, starting with vehicle and person actions.
It can tell whether a car is broken down or someone is busy unloading a truck. It should also be able to identify whether a human is jogging, jumping in place, standing around talking on their phone, or waiting to cross the road, both legally and illegally. It can even tell the age of the human and how they might react in a situation. The most obvious example is a child running into the road after a ball, which the vehicle, in theory, should be able to predict.
It's an exciting patent and another step in the right direction to reach Level 5 autonomous driving. Once Honda reaches what it believes to be human-level intelligence (not that hard these days), it should also be fine taking responsibility for the car's actions, which seems to be a problem for every automaker except Mercedes-Benz.
As for the inevitable debate about privacy, Honda will have to ask permission before accessing data on this scale. We're willing to bet early adopters would be happy to sign up for this system, which is much better than the Tesla approach of using the entire world as your own personal crash-testing facility.
Join The Discussion