Tesla’s head of artificial intelligence has revealed new footage showing its auto labelling tool for its self-driving initiative. It’s believed to provide a significant boost to Tesla’s Full Self-Driving Beta.
Tesla is frequently claimed to have a significant lead in self-driving data since it equipped all of its cars with sensors and has collected real-world data from a fleet of over a million vehicles. The company can use the massive data collection to develop the neural nets that power its suite of Autopilot capabilities, which it says will eventually lead to complete self-driving capability.
Furthermore, when such data is “labelled,” it means that the information in the photographs collected by the fleet is tagged with info like vehicles, lanes, street signs, and other nearby objects to create a panoptic awareness of the vehicle’s environment.
Elon Musk’s company is one step closer to level-3 automation thanks to Tesla’s self-labelling AI. However, Tesla is still leaving a lot of valuable data on the table, even with thousands of employees manually labelling videos. Developing an auto-labelling system that can automatically and reliably classify enormous amounts of footage is the holy grail of labelling.
Tesla has indicated that it is working on such a tool, which will work in harmony with the Dojo supercomputer. According to Andrej Karpathy, the firm’s Senior Director of AI, the organisation has made significant progress. The new tweets contained images and video feeds from Tesla’s auto-labelling technology.
“Some panoptic segmentation eye candy from a new project we are bringing up,” wrote Karpathy in his tweet.
“These are too raw to run in the car but feed into auto labellers. Collaboration of data labelling a large (100K+), clean, diverse, multicam+video dataset and engineers who train the models.”
“The multicam + video data, temporal continuity of a slowly moving viewpoint, close collaboration with data sourcing and labelling, and the infinity-sized dataset of unlabeled clips dramatically expands creative modelling opportunities on the neural net side,” added Karpathy in a tweet.
Moreover, Karpathy said that it’s still early in the deployment of this technology, and he appears to be sharing the footage to recruit more people for his team.
Although the new developments are thrilling, it’s essential remembering that Tesla has admitted that Elon Musk exaggerates when it comes to “Full Self-Driving.” In addition, the legal transparency organisation PlainSite published a letter this year that indicates a gap between what Musk says about Tesla’s Autopilot and the actual capabilities of the underlying software.
“Elon’s tweet does not match engineering reality per CJ,” said the memo from California’s DMV, referring to its March 9 conference call with officials at Tesla.
“Tesla is at Level 2 currently.”
However, given the advancement of new auto-labelling technologies, making real-world forecasts is not out of the question. Tesla vehicles will have level-3 automation in the future.
