PTAB

IPR2025-01573

Tesla Inc v. Perceptive Automata LLC

Log In
Key Events
Petition
petition

1. Case Identification

2. Patent Overview

  • Title: Predicting Road User Responses for Autonomous Vehicle Control
  • Brief Description: The ’344 patent discloses a computer-implemented method for controlling an autonomous vehicle. The system uses a supervised learning model that is trained with data reflecting human responses to various road scenarios. This trained model then receives live sensor data, predicts an output statistical summary characterizing how a plurality of users would likely respond, and controls the vehicle’s operation based on that summary.

3. Grounds for Unpatentability

Ground 1: Claims 1-22 are obvious over Ross in view of Cox

  • Prior Art Relied Upon: Ross (Patent 10,496,091) and Cox (International Publication No. WO 2014/210334).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Ross taught most limitations of the challenged claims. Ross disclosed an autonomous vehicle system that used a trained "intent model" to predict the actions and trajectories of other road users based on training data generated by human operators. To supply the remaining limitations related to a "supervised learning model" that predicts "user responses," Petitioner combined Ross with Cox. Cox taught augmenting machine learning with crowd-sourced data from human populations to train supervised models that "better mimic human performance" and explicitly suggested applying this technique to "driverless" automobile applications.
    • Motivation to Combine: A Person of Ordinary Skill in the Art (POSITA) would combine Cox's human-centric training methodology with Ross's intent model to achieve Ross's stated goal of creating a vehicle that functions in a more "human-like" or "polite" way. Cox provided a known method for improving predictive models to better reflect human decision-making, which represented a natural and predictable improvement to Ross's system.
    • Expectation of Success: A POSITA would have had a high expectation of success because the combination involved applying a known training technique (Cox) to a known type of system (Ross's predictive model) to achieve the predictable result of a more human-like predictive output.

Ground 2: Claims 1-22 are obvious over Ross, Cox, and Ogale

  • Prior Art Relied Upon: Ross (Patent 10,496,091), Cox (WO 2014/210334), and Ogale (Patent 10,733,506).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground was presented as an alternative to Ground 1. To the extent the Ross and Cox combination was found not to explicitly teach that the model’s output "predicts how a human would have labeled the input sensor data," Petitioner argued Ogale supplied this teaching. Ogale, another patent related to autonomous vehicles, explicitly disclosed a neural network trained with data from "human participants" where the model’s prediction "predicts how a human would have labeled the input sensor data."
    • Motivation to Combine: A POSITA implementing the system of Ross and Cox would have looked to a reference like Ogale for further implementation details. Ogale provided an explicit solution for making a predictive model's output directly correspond to human responses, which would improve the accuracy and safety of the Ross-Cox system. All three references were directed to improving predictive models for autonomous vehicles using human-annotated data, making the combination logical.
    • Expectation of Success: Success was predictable because Ogale merely made explicit a desired outcome of the Ross and Cox combination, providing known techniques to achieve it with minimal modification.

Ground 3: Claims 1-4, 7-13, and 16-22 are obvious over Zhang

  • Prior Art Relied Upon: Zhang (Application # 2019/0072966).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner contended that Zhang, as a single reference, rendered the challenged claims obvious. Zhang disclosed a "prediction-based system" for an autonomous vehicle that used a trained neural network (a supervised learning model) to analyze camera data. The model was trained using "real world, human behavior data" that was labeled by humans. In operation, Zhang’s system executed the model to generate a "set of predicted trajectories and confidence levels" for objects in the environment, which Petitioner asserted was a "statistical summary." The autonomous vehicle was then controlled based on this output to avoid potential conflicts.
    • Motivation to Combine: This was not a combination ground. Petitioner argued that Zhang's disclosure of a system trained on human-labeled driving behaviors to predict future trajectories inherently taught a model that predicted a distribution of human responses.
    • Expectation of Success: Not applicable for a single-reference obviousness ground; the argument was that Zhang’s disclosed system inherently met the claim limitations.

4. Relief Requested

  • Petitioner requested institution of an inter partes review and cancellation of claims 1-22 of Patent 10,614,344 as unpatentable.