DCT

7:25-cv-00594

Perceptive Automata LLC v. Tesla Inc

Log In
Key Events
Complaint

I. Executive Summary and Procedural Information

  • Parties & Counsel:
  • Case Identification: 7:25-cv-00594, W.D. Tex., 12/29/2025
  • Venue Allegations: Plaintiff alleges venue is proper in the Western District of Texas because Defendant Tesla, Inc. maintains its global headquarters, a major manufacturing facility ("Gigafactory"), and other regular and established places of business within the district, and has allegedly committed acts of infringement there.
  • Core Dispute: Plaintiff alleges that Defendant’s Full Self-Driving (FSD) autonomous vehicle technology infringes five patents related to systems and methods for predicting human behavior and interactions with vehicles using machine learning models trained on human observational data.
  • Technical Context: The technology at issue addresses the challenge of enabling autonomous vehicles to navigate complex urban environments by predicting the intentions and "state of mind" of human road users, such as pedestrians and other drivers.
  • Key Procedural History: The complaint notes prior litigation between the parties in the Eastern District of Texas (the "EDTX Case"), which may be relevant to Defendant's knowledge of the asserted patents. It also alleges that Defendant was aware of the patents through citations made during the prosecution of its own patent applications.

Case Timeline

Date Event
2017-07-05 Priority Date for ’344, ’889, and ’046 Patents
2017-11-01 Waymo begins operating fully self-driving vehicles without a safety driver
2019-01-30 Priority Date for ’346 Patent
2019-02-06 Priority Date for ’579 Patent
2020-04-07 U.S. Patent No. 10,614,344 Issues
2020-10-01 Tesla releases beta version of its FSD software
2021-09-21 U.S. Patent No. 11,126,889 Issues
2022-10-11 U.S. Patent No. 11,467,579 Issues
2022-12-06 U.S. Patent No. 11,520,346 Issues
2023-09-12 U.S. Patent No. 11,753,046 Issues
2024-01-01 Tesla re-incorporates from Delaware to Texas
2025-12-29 Complaint Filed

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 10,614,344 - "System and Method of Predicting Human Interaction with Vehicles"

  • Patent Identification: U.S. Patent No. 10,614,344, "System and Method of Predicting Human Interaction with Vehicles," issued April 7, 2020.

The Invention Explained

  • Problem Addressed: The patent's background section identifies that computers and autonomous vehicles cannot adequately predict the behavior of people, particularly in complex urban environments, and that existing methods relying on simple "motion vectors" yield "inferior results." (’344 Patent, col. 1:35-49).
  • The Patented Solution: The invention proposes a system that collects sensor data (e.g., video) from a vehicle, presents it as "stimulus" to a plurality of human observers, and collects their predictions on how a person in the scene will act. (’344 Patent, col. 5:39-46). The system then generates "summary statistics" based on these collective human responses and uses this statistical data to train a supervised learning model. (’344 Patent, col. 6:1-11). This trained model is then used in the autonomous vehicle to receive live sensor data, predict a distribution of expected human responses, and use that prediction to control the vehicle's operation. (’344 Patent, col. 7:5-39; Fig. 2A).
  • Technical Importance: This approach aims to create a more sophisticated prediction model that moves beyond simple physics-based extrapolation to incorporate the nuanced, predictive "intuition" of human drivers. (Compl. ¶51).

Key Claims at a Glance

  • The complaint asserts independent claims 1 (a method) and 19 (a computing system). (Compl. ¶22).
  • Independent Claim 1 includes the core steps of:
    • receiving sensor data from an autonomous vehicle displaying an object;
    • providing the sensor data as input to a supervised learning model configured to predict an "output statistical summary characterizing a distribution of user responses";
    • executing the trained model to generate the statistical summary data based on new sensor data; and
    • controlling the vehicle's operation based on the generated statistical summary data.
  • Independent Claim 19 recites a computing system comprising a processor and memory configured to perform the steps of the method claim.
  • The complaint expressly reserves the right to assert additional claims. (Compl. ¶22, n.1).

U.S. Patent No. 11,126,889 - "Machine Learning Based Prediction of Human Interactions with Autonomous Vehicles"

  • Patent Identification: U.S. Patent No. 11,126,889, "Machine Learning Based Prediction of Human Interactions with Autonomous Vehicles," issued September 21, 2021.

The Invention Explained

  • Problem Addressed: As a continuation of the same patent family, the ’889 Patent addresses the same core problem: the inability of autonomous vehicles to adequately predict human behavior in urban environments. (’889 Patent, col. 1:39-53).
  • The Patented Solution: This patent claims a system and method that focuses specifically on predicting the "state of mind" of road users. The system involves receiving a plurality of images, receiving "a plurality of user responses, each user response describing a state of mind of a road user displayed in one or more images," generating a training dataset from these responses, training a model, and then using that model in an autonomous vehicle to predict a road user's state of mind from new images to control the vehicle. (’889 Patent, Abstract; col. 15:1-16:21).
  • Technical Importance: The invention refines the predictive model by explicitly targeting a road user's "state of mind"—such as intent, awareness, or disposition—as the key variable for improving autonomous navigation. (Compl. ¶¶ 91-92).

Key Claims at a Glance

  • The complaint asserts independent claims 1 (a computer system) and 2 (a computer-implemented method). (Compl. ¶84).
  • Independent Claim 2 includes the core steps of:
    • receiving a plurality of images of road scenes;
    • receiving a plurality of user responses, each describing a "state of mind" of a road user in an image;
    • generating a training dataset from these responses;
    • training a supervised learning model with the dataset;
    • in an autonomous vehicle, receiving a new image with a road user;
    • predicting, using the model, statistics describing the "state of mind" of the road user in the new image; and
    • controlling the vehicle based on the prediction.
  • Independent claim 1 recites a computer system configured to perform these method steps.
  • The complaint expressly reserves the right to assert additional claims. (Compl. ¶22, n.1).

U.S. Patent No. 11,467,579 - "Probabilistic Neural Network for Predicting Hidden Context of Traffic Entities for Autonomous Vehicles"

  • Patent Identification: U.S. Patent No. 11,467,579, "Probabilistic Neural Network for Predicting Hidden Context of Traffic Entities for Autonomous Vehicles," issued October 11, 2022.
  • Technology Synopsis: This patent describes a method for navigating autonomous vehicles by first training a "probabilistic neural network" to generate output representing "hidden context" (e.g., intent, awareness) for a traffic entity. The system then uses this trained network in a vehicle to execute on new images, determine a "measure of uncertainty" for the predicted context, and navigate the vehicle based on that measure of uncertainty. (’579 Patent, Abstract; col. 17:180-18:189).
  • Asserted Claims: Independent claims 1 (method) and 20 (computer system). (Compl. ¶148).
  • Accused Features: The complaint alleges that Tesla's FSD software, which it characterizes as a probabilistic neural network, infringes by generating feature vectors and outputs representing hidden context to navigate its vehicles. (Compl. ¶¶ 154, 157, 161).

U.S. Patent No. 11,520,346 - "Navigating Autonomous Vehicles Based on Modulation of a World Model Representing Traffic Entities"

  • Patent Identification: U.S. Patent No. 11,520,346, "Navigating Autonomous Vehicles Based on Modulation of a World Model Representing Traffic Entities," issued December 6, 2022.
  • Technology Synopsis: This invention involves generating a point cloud representation of a vehicle's surroundings, identifying a traffic entity, using a machine learning model to determine the entity's "hidden context," and then "modifying" a predicted region of travel for that entity based on the determined context. The autonomous vehicle is then navigated to maintain a threshold distance from this modified region. (’346 Patent, Abstract; col. 1:34-67).
  • Asserted Claims: Independent claims 1 (method) and 19 (computer system). (Compl. ¶209).
  • Accused Features: The complaint alleges infringement by Tesla's FSD software, which is accused of identifying traffic entities, determining their projected paths, determining a hidden context via a machine learning model, and modifying the vehicle's navigation based on that context. (Compl. ¶¶ 218-223).

U.S. Patent No. 11,753,046 - "System and Method of Predicting Human Interaction with Vehicles"

  • Patent Identification: U.S. Patent No. 11,753,046, "System and Method of Predicting Human Interaction with Vehicles," issued September 12, 2023.
  • Technology Synopsis: This patent, from the same family as the ’344 and ’889 patents, claims a method focused on the training process. It involves storing images, sending them to human observers with a question about a user's "state of mind," receiving responses, generating summary statistics, storing those statistics with the image as training data, and then training a model with that data to predict the state of mind in a new image. (’046 Patent, Abstract; col. 15:303-16:303).
  • Asserted Claims: Independent claims 1 (method) and 15 (computing system). (Compl. ¶271).
  • Accused Features: The complaint accuses Tesla's "Data Engine" and its use of over 1,000 in-house data labelers to generate training data for its FSD models, alleging this process maps onto the claimed training method. (Compl. ¶¶ 276-281).

III. The Accused Instrumentality

Product Identification

The accused instrumentalities are Tesla vehicles equipped with Full Self-Driving (FSD) software and hardware (e.g., Model Y, Cybertruck) and the supporting datacenters (e.g., Dojo, Cortex) used to train the FSD models. (Compl. ¶¶ 15-16, 22).

Functionality and Market Context

The FSD system is an advanced driver-assistance system that uses an array of cameras and an onboard computer to navigate the vehicle. (Compl. ¶¶ 28, 57). The complaint alleges that the FSD software is a "supervised learning based model" that is trained on video data collected from Tesla's vehicle fleet. (Compl. ¶¶ 29, 90). This training process allegedly involves a team of over 1,000 human "data labelers" who provide "advanced labeling" that describes the state of mind or context of road users in the collected video clips. (Compl. ¶¶ 29, 91). The complaint points to a Tesla AI Day presentation showing a "Data Engine" diagram as evidence of this training and data collection process. (Compl. ¶90; p. 58). The trained FSD model then runs in the vehicle to make real-time predictions about the behavior of other road users and control the vehicle accordingly. (Compl. ¶¶ 31, 96).

IV. Analysis of Infringement Allegations

’344 Patent Infringement Allegations

Claim Element (from Independent Claim 1) Alleged Infringing Functionality Complaint Citation Patent Citation
receiving, by a computing device... sensor data captured by a sensor installed on the autonomous vehicle, the sensor data displaying an object on the road The FSD hardware and software on a Tesla vehicle function as the computing device, receiving image data from the vehicle's cameras which displays objects like other cars and pedestrians. ¶28 col. 5:4-7
providing the sensor data as input to a supervised learning based model, the supervised learning based model configured to receive an input sensor data... and predicting an output statistical summary characterizing a distribution of user responses... The FSD software is alleged to be a supervised learning model that receives the camera data and predicts outcomes. The complaint alleges this prediction corresponds to a distribution of user responses, citing FSD's creation of "goal candidates" derived from "human demonstration." ¶29 col. 5:8-15
executing the trained supervised learning based model to generate a statistical summary data characterizing a distribution of user responses... The onboard FSD software executes the trained model to generate this predictive output, which the complaint alleges is evidenced by the FSD software's "Intervention Likelihood" and "Human-like Discriminator" scoring mechanisms. ¶30 col. 6:1-2
controlling the operation of the autonomous vehicle on the road based on the generated statistical summary data The FSD system uses the model's output to control the vehicle, as illustrated in a diagram from a Tesla presentation showing the system deciding whether to "Assert or Yield to the pedestrian." ¶31; p. 24 col. 7:35-39
  • Identified Points of Contention:
    • Scope Questions: A primary question may be whether Tesla's internal team of over 1,000 data labelers who annotate historical driving data qualifies as the "plurality of users" providing "responses" as contemplated by the patent. The defense may argue this is a distinct process from the patent's description of presenting "stimulus" to "human observers" for predictive feedback.
    • Technical Questions: The analysis will question whether the output of Tesla's neural network, described with terms like "Intervention Likelihood," is technically equivalent to the claimed "statistical summary characterizing a distribution of user responses." The complaint alleges this equivalence, but the actual implementation of the FSD software's output will be a key factual issue.

’889 Patent Infringement Allegations

Claim Element (from Independent Claim 2) Alleged Infringing Functionality Complaint Citation Patent Citation
receiving a plurality of images displaying road scenes captured by one or more vehicles Tesla's datacenters (e.g., Dojo, Cortex) receive millions of video clips from its fleet of vehicles operating on public roads. ¶90 col. 5:12-15
receiving a plurality of user responses, each user response describing a state of mind of a road user displayed in one or more images Tesla's in-house team of data labelers provides "advanced labeling" for the received video clips, choosing from options like "stopped_traffic" or "parked," which the complaint alleges describes a state of mind. ¶91; p. 59 col. 5:46-53
generating a training dataset comprising summary statistics of user responses describing the state of minds of road users... At its datacenters, Tesla allegedly generates a training dataset by aggregating the labelers' responses, modifying statistical weights with each new response to improve model accuracy. ¶92 col. 6:8-9
training, using the training dataset, a supervised learning based model configured to predict summary statistics describing a state of mind... Tesla uses this curated training dataset to train the FSD software, which is alleged to be the claimed supervised learning model. ¶93 col. 6:44-47
receiving, by an autonomous vehicle, a new image captured by a camera of the autonomous vehicle, the new image of a scene including a road user A Tesla vehicle operating with FSD continuously receives new images from its cameras, capturing scenes that include road users like pedestrians. A screenshot of a Tesla vehicle approaching a pedestrian is provided as evidence. ¶94; p. 64 col. 7:12-15
predicting, by the autonomous vehicle, using the supervised learning based model, summary statistics describing a state of mind of the road user in the new image The FSD software allegedly uses the trained model to make real-time statistical predictions about the state of mind of road users, such as whether a car will turn or a pedestrian will cross the road. ¶95 col. 7:26-29
controlling the autonomous vehicle based on the prediction of the supervised learning based model The FSD software controls the vehicle based on the prediction, for example, by stopping to allow a pedestrian to cross the road without human intervention. ¶96 col. 7:41-44
  • Identified Points of Contention:
    • Scope Questions: The case may turn on the construction of "state of mind." The defense could argue that selecting a label like "stopped_traffic" is a factual annotation of a vehicle's status, not a description of the driver's mental state as required by the claim.
    • Technical Questions: A key question will be whether the steps of receiving images and user responses for training are performed as part of the claimed method for "controlling an autonomous vehicle." The defense may argue that the training steps are separate from the real-time control method that is practiced by the end-user vehicle.

V. Key Claim Terms for Construction

  • The Term: "user responses" (’344 Patent) / "user response describing a state of mind" (’889 Patent)

  • Context and Importance: The definition of this term is critical to determining whether Tesla's internal data labeling process infringes. Practitioners may focus on whether "users" can be internal employees performing a labeling task and whether selecting from a predefined list of labels (e.g., "stopped_traffic") constitutes a "response describing a state of mind."

  • Intrinsic Evidence for Interpretation:

    • Evidence for a Broader Interpretation: The specification describes collecting predictions from a "large number of users (or human observers)" to train an algorithm. (’344 Patent, col. 5:39-46). This language could support a broad definition that includes any group of humans providing input to a training dataset, including internal employees.
    • Evidence for a Narrower Interpretation: The specification describes a process where "stimulus is transmitted from the server 106 and displayed to... the user terminal 108" and observers "input their responses." (’344 Patent, col. 5:39-46). This could be interpreted more narrowly to imply an interactive, survey-like system presented to end-users or a test group, rather than an offline, internal data annotation process.

  • The Term: "statistical summary characterizing a distribution of user responses" (’344 Patent)

  • Context and Importance: Infringement of the ’344 Patent hinges on whether the output generated by the FSD software is properly characterized by this term. Practitioners may focus on the technical distinction between the output of a complex neural network and a direct statistical summary.

  • Intrinsic Evidence for Interpretation:

    • Evidence for a Broader Interpretation: The patent abstract describes creating a model "based on the statistical data" and applying it to predict user behavior. This suggests the output is functionally tied to the initial statistics, potentially allowing for a broader interpretation of any probabilistic output derived from such data.
    • Evidence for a Narrower Interpretation: The specification provides a specific example of statistics that "can be categorized in terms of how many human observers believe that the pedestrian will stop upon reaching the intersection." (’344 Patent, col. 6:6-9). This language could support a narrower definition requiring a direct aggregation of discrete human judgments, which may differ from the more abstract vector-space outputs of a modern neural network.

VI. Other Allegations

  • Indirect Infringement: The complaint alleges inducement of infringement based on Defendant's extensive customer support, instructions, demo drives, website marketing, and an internal email from Elon Musk requiring employees to "install and activate FSD" for customers on test rides, all of which allegedly instruct and encourage customers to use the FSD system in an infringing manner. (Compl. ¶¶ 77, 141, 202, 264, 327).
  • Willful Infringement: The complaint alleges willful infringement based on Defendant's alleged knowledge of the asserted patents. This knowledge is purportedly established through prior litigation between the parties (the "EDTX Case"), service of infringement contentions in that case, and numerous instances where Defendant allegedly cited Plaintiff's patents during the prosecution of its own patent portfolio before the USPTO. (Compl. ¶¶ 75, 80, 139, 144).

VII. Analyst’s Conclusion: Key Questions for the Case

  • A core issue will be one of definitional scope: can Tesla’s internal, professional data-labeling workforce be construed as the "plurality of users" providing "responses" as contemplated by the patents, which describe a system for presenting "stimuli" to "human observers" for predictive feedback?
  • A central evidentiary question will be one of functional mapping: does the abstract, probabilistic output of Tesla's complex neural network (described in the complaint with terms like "Intervention Likelihood") perform the specific functions required by the claims, such as generating a "statistical summary characterizing a distribution of user responses" or predicting a "state of mind," or is there a fundamental mismatch in technical operation?
  • A key legal question will be one of divided infringement: for the method claims that include both training steps (allegedly performed by Tesla at its datacenters) and operational steps (performed by the vehicle), the court will need to analyze whether Tesla directs or controls the performance of all steps of the claimed invention sufficient to establish direct infringement.