Tesla Inc v. Perceptive Automata LLC

1. Case Identification

• Case #: IPR2025-01576
• Patent #: 11,467,579
• Filed: October 9, 2025
• Petitioner(s): Tesla, Inc.
• Patent Owner(s): Perceptive Automata LLC
• Challenged Claims: 1-20

2. Patent Overview

• Title: Navigating Autonomous Vehicles Using Probabilistic Neural Networks
• Brief Description: The ’579 patent relates to a method for autonomous vehicle navigation that involves training a probabilistic neural network (PNN) to generate an output representing "hidden context" for a traffic entity, determining a measure of uncertainty for that output, and navigating the vehicle based on the determined uncertainty.

3. Grounds for Unpatentability

Ground 1: Claims 1-20 are obvious over Djuric in view of AA

• Prior Art Relied Upon: Djuric (Application # 2019/0049970) and AA (Application # 2020/0160535).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Djuric, which is directed to predicting the motion of objects for an autonomous vehicle, discloses most limitations of the challenged claims. Djuric teaches using a neural network trained with manually labeled data to generate multiple predicted trajectories for objects, calculating confidence levels (a measure of uncertainty) for those predictions, and using a motion planning system to navigate the vehicle based on this output. Petitioner asserted that AA supplies the explicit disclosure of a "probabilistic" neural network and the generation of a "feature vector," which Djuric describes in concept but not by name.
  • Motivation to Combine: A POSITA would combine AA with Djuric to improve Djuric’s prediction model. Djuric teaches that its model may be a Bayesian network, and Petitioner argued that AA’s disclosure of a probabilistic model provides more specific, known implementation details for improving Djuric’s ability to "account for the uncertainty of future movements." Incorporating AA’s feature vector generation would also be a predictable way to improve the computational efficiency of Djuric’s system.
  • Expectation of Success: Petitioner contended that a POSITA would have had a reasonable expectation of success, as the combination involves applying well-known techniques (a probabilistic framework and feature vectors) to a known autonomous vehicle prediction system (Djuric) to achieve the predictable results of improved uncertainty handling and efficiency.

Ground 2: Claims 1-20 are obvious over Djuric, AA, and Cox

• Prior Art Relied Upon: Djuric (Application # 2019/0049970), AA (Application # 2020/0160535), and Cox (WO 2014/210334).
• Core Argument for this Ground:
  • Prior Art Mapping: This ground builds on the Djuric and AA combination, adding Cox to explicitly teach the limitation of generating an output representing the "likelihood of receiving a particular user response." Petitioner highlighted that Cox, which shares an inventor with the ’579 patent, teaches a machine learning system trained with crowd-sourced data from human annotators. The goal of Cox is to make predictions that are "more consistent with the decisions of the human annotators" and "better mimic human performance," which Petitioner argued directly teaches training a model to generate outputs that reflect the likelihood of human responses.
  • Motivation to Combine: A POSITA would combine Cox with the Djuric/AA system to improve the quality of the training and the accuracy of the predictions. By using Cox's human-centric training methodology, the resulting prediction model would generate trajectory confidences that better reflect human intuition and decision-making, a stated goal of such systems. Cox expressly suggests its invention is applicable to "driverless/semiautonomous automobile" applications.
  • Expectation of Success: Success would be expected because it involves a straightforward application of an improved training data methodology (Cox) to a known type of prediction model (the Djuric/AA combination) to enhance its performance in a predictable manner.

Ground 3: Claims 1-20 are obvious over Djuric, AA, and Ross

• Prior Art Relied Upon: Djuric (Application # 2019/0049970), AA (Application # 2020/0160535), and Ross (Patent 10,496,091).

• Core Argument for this Ground:

  • Prior Art Mapping: This ground uses Ross to supplement the teachings of Djuric and AA, particularly for dependent claims reciting that the "hidden context" represents a "state of mind" (claim 7) or "degree of awareness" (claim 9). Petitioner asserted that while Djuric teaches "goal-based prediction," Ross explicitly discloses a system for "reasoning about the intents of objects, such as other road users" to allow an autonomous vehicle to function in a more "human-like" or "polite" way. Ross's disclosure of predicting a pedestrian's intent to cross a road directly teaches representing the "state of mind" of a traffic entity.
  • Motivation to Combine: A POSITA would have found it obvious to supplement Djuric's goal-based predictions with Ross's more nuanced intent-based reasoning. This would expand the predictive capabilities of the Djuric/AA model, leading to more accurate and safer navigation in complex scenarios involving interactions with human drivers and pedestrians.
  • Expectation of Success: The combination represents a logical improvement of a general prediction model (Djuric/AA) with a more specialized, known technique for intent prediction (Ross), which would have been well within the skill of a POSITA.

• Additional Grounds: Petitioner asserted further obviousness challenges (Grounds 3-4 and 6) adding Rolfe (Application # 2018/0247200) to confirm the well-known nature of using an Evidence Lower Bound (ELBO) for model optimization, and also presented combinations including both Cox and Ross.

4. Key Technical Contentions (Beyond Claim Construction)

• Printed Matter Doctrine: Petitioner argued that numerous claim limitations are directed to patent-ineligible printed matter and should be afforded no patentable weight. These limitations include those defining the content of information, such as "an image of traffic, the image displaying a traffic entity" (1[a-i]), "output representing hidden context" (1[a-iii]), and what the hidden context represents in claims 7-10 (e.g., "a state of mind"). Petitioner contended these limitations merely describe the information itself, rather than a functional relationship with the substrate (e.g., computer memory), and thus cannot be used to distinguish the claims from the prior art.

5. Relief Requested

• Petitioner requests institution of an inter partes review and cancellation of claims 1-20 of Patent 11,467,579 as unpatentable.