PTAB

IPR2025-01574

Tesla Inc v. Perceptive Automata LLC

Log In
Key Events
Petition
petition

1. Case Identification

2. Patent Overview

  • Title: Controlling an Autonomous Vehicle Based on Predicted State of Mind of Road Users
  • Brief Description: The ’889 patent relates to a system for controlling an autonomous vehicle by using a supervised machine learning model. The model is trained on a dataset of human responses describing the "state of mind" of road users in images, and is then used to predict the state of mind of road users in new, live images to inform vehicle control.

3. Grounds for Unpatentability

Ground 1: Claims 1-20 are obvious over Huval in view of Ross.

  • Prior Art Relied Upon: Huval (Application # 2018/0373980) and Ross (Patent 10,496,091).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Huval taught most limitations of the challenged claims, including an autonomous vehicle system that uses a machine learning model (e.g., a neural network) trained with human-labeled image data to detect objects and control the vehicle. However, Huval was primarily focused on object detection. Ross allegedly supplied the missing elements by teaching an "intent model" for an autonomous vehicle trained with human-influenced data to predict the intent and next actions of road users (e.g., pedestrians, cyclists). Ross explicitly taught controlling the vehicle based on these intent predictions, such as a pedestrian's intent "to cross the road when clear."
    • Motivation to Combine: Petitioner contended a POSITA would combine Huval and Ross because Ross provided a known application for Huval's object detection system: predicting behavior and intent. A POSITA would be motivated to enhance Huval's object detection capabilities with Ross's intent prediction to create a safer, more robust, and more "human-like" autonomous driving system capable of anticipating the actions of other road users to avoid collisions. Ross’s teachings were presented as a natural and beneficial extension of Huval's system.
    • Expectation of Success: A POSITA would have had a reasonable expectation of success because the combination involved applying the known technique of intent prediction (from Ross) to a conventional autonomous vehicle perception system (from Huval) to achieve the predictable result of improved environmental awareness.

Ground 2: Claims 1-20 are obvious over Huval, Ross, and Cox.

  • Prior Art Relied Upon: Huval (Application # 2018/0373980), Ross (Patent 10,496,091), and Cox (WO 2014/210334).

  • Core Argument for this Ground:

    • Prior Art Mapping: This ground built upon the Huval-Ross combination, adding Cox to explicitly teach limitations related to generating and using "summary statistics" for model training. Petitioner highlighted that Cox, which shared an inventor with the ’889 patent, taught augmenting machine learning with measurements from human populations. Cox described collecting crowd-sourced training data, compiling statistical information about the human responses (including psychometric data like response times and accuracy), and using this statistical data to train a supervised learning model to "better mimic human performance." This directly mapped to claim elements requiring a training dataset comprising summary statistics of user responses.
    • Motivation to Combine: Petitioner argued a POSITA would be motivated to incorporate Cox’s teachings into the Huval-Ross framework to implement well-known and routine methods for optimizing model training. Cox provided specific implementation details for generating and using statistical and psychometric data from annotator responses to improve the accuracy of the predictive model taught by Huval and Ross. The significant technical overlap between all three references—all being directed to machine learning for autonomous systems—would have prompted a POSITA to look to Cox for established training optimization techniques.
    • Expectation of Success: A POSITA would have reasonably expected success in applying Cox's detailed statistical training methods to the established Huval-Ross system, as it represented a predictable use of known data analysis techniques to improve the performance of a known type of predictive model.

  • Additional Grounds: Petitioner asserted additional obviousness challenges over combinations including Munro (Application # 2016/0162456), which taught tallying annotator responses to create summary statistics. These grounds relied on similar theories of combining known machine learning training and data aggregation techniques to arrive at the claimed invention.

4. Relief Requested

  • Petitioner requests institution of IPR and cancellation of claims 1-20 of Patent 11,126,889 as unpatentable.