Tesla Inc v. Perceptive Automata LLC

1. Case Identification

• Case #: IPR2025-01575
• Patent #: 11,753,046
• Filed: October 1, 2025
• Petitioner(s): Tesla, Inc.
• Patent Owner(s): Perceptive Automata LLC
• Challenged Claims: 1-19

2. Patent Overview

• Title: Machine Learning Models for Predicting User State of Mind
• Brief Description: The ’046 patent relates to a computer-implemented method for training a supervised machine learning model. The method involves generating training data by collecting judgments from multiple human observers about the "state of mind" of a user depicted in an image, using this data to train a model, and then executing the model to predict the state of mind of a user in a new image.

3. Grounds for Unpatentability

Ground 1: Obviousness over a Single Reference - Claims 1-19 are obvious over Cox.

• Prior Art Relied Upon: Cox (WO 2014/210334).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Cox, which shares a common inventor with the ’046 patent, discloses all limitations of the challenged claims. Cox teaches a machine learning system that uses crowd-sourced data from human annotators to train a model to "better mimic human performance." Specifically, Cox discloses presenting images of people to human annotators and asking them to select a corresponding emotion (e.g., jealous, panicked), which Petitioner contended is a "state of mind." Cox further teaches receiving the annotators' responses, generating "psychometric data" and "item-response curves" (summary statistics) from these responses, storing this data in a database associated with the images, and using the data to train a classification model. Finally, Cox describes using the trained model to make predictions on new images.
  • Key Aspects: Petitioner emphasized that Cox is analogous art directed to the same problem of making more accurate model predictions using human-annotated training data, and that its own inventor failed to disclose this reference during prosecution.

Ground 2: Obviousness over Combined References - Claims 1-19 are obvious over Cox in view of Ross.

• Prior Art Relied Upon: Cox (WO 2014/210334) and Ross (Patent 10,496,091).
• Core Argument for this Ground:
  • Prior Art Mapping: This ground asserted that to the extent Cox’s teaching of predicting "emotion" is not considered a "state of mind," Ross provides this missing element. Ross discloses training an "intent model" for an autonomous vehicle to predict the intent of road users (e.g., pedestrians, other drivers) based on human-labeled training data. Petitioner argued that predicting a road user's intent to perform an action (e.g., cross the road, change lanes) is equivalent to predicting their "state of mind." The combination, therefore, explicitly taught asking human observers about a user’s likely actions and training a model to predict this intent.
  • Motivation to Combine: A POSITA would combine Cox and Ross because Ross provides a well-known and advantageous application (autonomous driving) for Cox’s general machine learning framework. Cox itself suggests applying its techniques to "driverless" vehicles. Both references aim to create models that produce more "human-like" predictions, providing a clear reason to integrate Ross's specific intent-prediction methods into Cox’s crowd-sourcing and training system.
  • Expectation of Success: A POSITA would have a reasonable expectation of success because combining the references involved applying Ross’s known techniques for labeling road user intent to Cox's established system for collecting and processing human annotations, which would predictably yield a model capable of predicting user intent.

Ground 3: Obviousness over Combined References - Claims 1-19 are obvious over Ellenbogen in view of Munro.

• Prior Art Relied Upon: Ellenbogen (Application # 2017/0099200) and Munro (Application # 2016/0162456).

• Core Argument for this Ground:

  • Prior Art Mapping: Petitioner argued that Ellenbogen discloses a system where images are sent to multiple "human agents" who are asked questions about characteristics of the image, such as a person's behavior (e.g., "suspicious behavior," "tailgating"). This was mapped to the claim limitations of generating data based on questions about a user's "state of mind." Munro was introduced to teach the specific generation and storage of "summary statistics." Munro discloses an "aggregation process" that involves tallying the number of annotators who classified a stimulus with a particular label. The combination taught generating a composite output of tallied human responses (per Munro) to questions about user behavior (per Ellenbogen) and using that aggregated statistical data to train a predictive model.
  • Motivation to Combine: A POSITA would combine these references because Munro provides routine implementation details for Ellenbogen’s system. Ellenbogen teaches generating a "composite output" from human agents, and Munro teaches a specific, well-known method (tallying) for creating such an output and using it as statistical training data. This combination would be a predictable enhancement to improve the accuracy and efficiency of Ellenbogen’s model training process.
  • Expectation of Success: The combination involved applying Munro's standard method for organizing annotator responses to Ellenbogen's system, a routine task that would predictably result in a more robust training dataset and improved model performance.

• Additional Grounds: Petitioner asserted additional obviousness challenges, including claims 5, 12, and 19 over Cox and Ellenbogen, and claims 1-19 over Cox and Munro. These grounds relied on similar theories where a secondary reference was used to supply a specific implementation of a broader concept taught in Cox.

4. Key Technical Contentions (Beyond Claim Construction)

• Printed Matter Doctrine: Petitioner argued that numerous claim limitations should be given no patentable weight under the printed matter doctrine. It contended that limitations reciting the content of information—such as an "image displaying one or more users," a "request to answer a question about a state of mind," or "summary statistics describing the state of mind"—are directed to the information itself, not to any patent-eligible structure. Petitioner asserted these limitations lack a functional relationship with the substrate (e.g., computer memory) and merely inform people of the information's content, making them patent-ineligible.

5. Relief Requested

• Petitioner requests institution of an inter partes review and cancellation of claims 1-19 of the ’046 patent as unpatentable under 35 U.S.C. §103.