PTAB

IPR2025-01577

Tesla Inc v. Perceptive Automata LLC

Log In
Key Events
Petition
petition

1. Case Identification

2. Patent Overview

  • Title: Autonomous Vehicle Navigation Using Machine Learning to Predict Traffic Entity Behavior
  • Brief Description: The ’346 patent describes a method for an autonomous vehicle to navigate by using sensor data, including images, to generate a point cloud representation of its surroundings. The system uses a machine learning model to determine a "hidden context," such as the intent of other traffic entities, modifies a predicted region of travel for those entities based on this context, and navigates to maintain a safe distance from the modified region.

3. Grounds for Unpatentability

Ground 1: Obviousness over Djuric - Claims 1-23 are obvious over Djuric.

  • Prior Art Relied Upon: Djuric (Application # 2019/0049970).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Djuric disclosed all key elements of the challenged claims. Djuric taught an autonomous vehicle with cameras and LiDAR sensors that capture image and point cloud data to identify surrounding objects. The system used a "machine-learned model" to make "goal-based predictions" of future trajectories for these objects, which Petitioner contended corresponded to the ’346 patent's "hidden context." Djuric’s model outputted multiple predicted trajectories with confidence levels, and the system selected the most likely one (thereby "modifying the region" of potential travel) to generate a motion plan for controlling the vehicle to avoid objects.
    • Motivation to Combine (for §103 grounds): This ground did not involve a combination of references. However, Petitioner argued it would have been obvious for a person of ordinary skill in the art (POSITA) to adapt Djuric's collision avoidance system to navigate the vehicle to stay at least a threshold distance away from the predicted path, as this is a fundamental and predictable goal of any autonomous navigation system.
    • Expectation of Success (for §103 grounds): A POSITA would have a reasonable expectation of success in implementing basic collision avoidance logic, as Djuric’s system was already designed to ensure its vehicle does not collide with objects.

Ground 2: Obviousness over Djuric in view of Cox - Claims 1-23 are obvious over Djuric and Cox.

  • Prior Art Relied Upon: Djuric (Application # 2019/0049970) and Cox (WO 2014/210334).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground asserted that to the extent Djuric did not explicitly teach a model outputting "summary statistics of expected human responses describing a hidden context," Cox supplied this teaching. Cox, which listed an inventor in common with the ’346 patent, disclosed a machine learning methodology that used crowd-sourced training data from human annotators to compile statistical information on human responses. The model was then trained on this statistical data to make predictions that "better mimic human performance" and are "more consistent with the decisions of the human annotators."
    • Motivation to Combine (for §103 grounds): A POSITA would combine Cox's human-centric training approach with Djuric's autonomous vehicle system to improve the system's predictive accuracy. By training Djuric's goal-based model with data reflecting human decisions as taught by Cox, the model would better predict the behavior of human drivers and pedestrians, thus enhancing safety. Cox explicitly suggested its techniques could be applied to "driverless / semiautonomous automobile" applications.
    • Expectation of Success (for §103 grounds): A POSITA would have a high expectation of success, as the combination involved applying a known training improvement technique (Cox) to a known type of system (Djuric) to achieve the predictable result of improved, more human-like predictions.

Ground 3: Obviousness over Djuric in view of Zhu - Claims 1-23 are obvious over Djuric and Zhu.

  • Prior Art Relied Upon: Djuric (Application # 2019/0049970) and Zhu (Application # 2019/0164018).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground argued that if Djuric was found insufficient in its teachings of generating and using a point cloud, Zhu provided the missing elements. Zhu taught combining LiDAR point cloud data with camera image data by projecting the 3D point cloud onto the 2D image data. This process created a detailed "drivable road surface map," which served as a point cloud representation of the surroundings upon which predicted trajectories could be determined and modified.
    • Motivation to Combine (for §103 grounds): A POSITA would combine Zhu's advanced mapping technique with Djuric's prediction system to improve overall navigation and object detection. Using Zhu's method to create a drivable road surface map in Djuric's system would allow for more accurate identification of drivable versus non-drivable areas, improving the safety and reliability of the vehicle’s motion planning.
    • Expectation of Success (for §103 grounds): Success was predictable because both references operated in the same technical field of autonomous vehicle perception. Integrating Zhu's sensor data fusion technique into Djuric's system was a straightforward application of known methods to enhance environmental modeling.
  • Additional Grounds: Petitioner asserted additional obviousness challenges over combinations including Ross (Patent 10,496,091), which was argued to teach reasoning about the "intents" of other road users to achieve more "human-like" or "polite" vehicle behavior, further specifying the nature of the "hidden context."

4. Key Technical Contentions (Beyond Claim Construction)

  • Printed Matter Doctrine: Petitioner argued that many claim limitations were not entitled to patentable weight under the printed matter doctrine. It contended that limitations merely reciting the informational content of data—such as "sensor data comprising one or more images," a "hidden context represents a state of mind," or a "point cloud representation of the surroundings"—lacked the required functional relationship with the substrate (e.g., computer memory) on which they exist. Petitioner asserted these limitations only inform people of the claimed information and could not be relied upon to distinguish the claims from the prior art.

5. Relief Requested

  • Petitioner requests institution of IPR and cancellation of claims 1-23 as unpatentable.