DCT

1:22-cv-01466

Autonomous Devices LLC v. Tesla Inc

Log In

I. Executive Summary and Procedural Information

  • Parties & Counsel:
  • Case Identification: 1:22-cv-01466, D. Del., 11/07/2022
  • Venue Allegations: Venue is alleged to be proper in the District of Delaware because Defendant Tesla, Inc. is a Delaware corporation.
  • Core Dispute: Plaintiff alleges that Defendant’s autonomous driving systems (Autopilot, FSD) and its AI training supercomputer (Dojo) infringe six patents related to machine learning, autonomous vehicle control, and simulation technology.
  • Technical Context: The technology at issue involves methods for training artificial intelligence systems to operate autonomous vehicles by learning from a fleet of vehicles and through simulated environments, a key area of development in the automotive industry.
  • Key Procedural History: The complaint does not reference any prior litigation, Inter Partes Review (IPR) proceedings, or licensing history related to the asserted patents.

Case Timeline

Date Event
2014-01-01 Tesla begins equipping Model S with Autopilot hardware (HW1)
2015-10-01 Initial release of Tesla Autopilot software
2016-02-16 Earliest Priority Date for ’585 Patent
2016-05-19 Earliest Priority Date for ’449 and ’583 Patents
2016-10-01 Tesla transitions to Enhanced Autopilot hardware (HW2)
2016-11-02 Earliest Priority Date for ’974 and ’344 Patents
2016-12-19 Earliest Priority Date for ’134 Patent
2017-08-01 Tesla announces hardware version 2.5 (HW2.5)
2018-10-05 Tesla releases Software Version 9.0; alleged infringement begins
2018-10-16 U.S. Patent No. 10,102,449 issues
2019-03-01 Tesla transitions to hardware version 3 (HW3)
2019-10-22 U.S. Patent No. 10,452,974 issues
2020-03-31 U.S. Patent No. 10,607,134 issues
2020-08-01 Tesla announces use of "Dojo" supercomputer for simulation training
2021-07-06 U.S. Patent No. 11,055,583 issues
2021-09-07 U.S. Patent No. 11,113,585 issues
2022-02-02 U.S. Patent No. 11,238,344 issues
2022-11-07 Complaint filed

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

U.S. Patent No. 10,452,974 - Artificially intelligent systems, devices, and methods for learning and/or using a device's circumstances for autonomous device operation

(Issued October 22, 2019)

The Invention Explained

  • Problem Addressed: The patent asserts that prior autonomous or semi-autonomous systems were limited because they relied on direct user commands or pre-coded, static instructions, and therefore "lack[ed] a way to learn [the] operation of a device or system and enable autonomous operation" based on experience (’974 Patent, col. 1:31-35; Compl. ¶43).
  • The Patented Solution: The invention proposes a system where a knowledgebase is built by creating correlations between a "circumstance representation" (e.g., sensor data identifying objects in the environment) and an instruction set for operating a device. This knowledge is acquired through a "learning process" that involves a user operating a "first device." Subsequently, a "second device" can autonomously operate by detecting a new circumstance, finding a partial match in the knowledgebase, and executing the correlated instruction set (’974 Patent, Abstract; col. 2:37-67; Compl. ¶44).
  • Technical Importance: This approach enables a system to learn from real-world operations and apply that learned knowledge to new situations, forming the basis for a fleet-learning model for autonomous vehicles (Compl. ¶17).

Key Claims at a Glance

  • The complaint asserts independent claims 1 and 18, and dependent claim 14 (Compl. ¶95).
  • Independent Claim 1 of the ’974 Patent recites essential elements including:
    • A memory that stores a knowledgebase containing a "first correlation" between a "first circumstance representation" and a "first one or more instruction sets."
    • The first correlation is "learned in a learning process that includes operating the first device at least partially by a user."
    • A "second device" detects a "third circumstance representation" using its sensors.
    • The system anticipates instruction sets for the second device based on "at least a partial match" between the third and first circumstance representations.
    • The second device "autonomously performs one or more operations defined by the first one or more instruction sets" in response to the anticipation.

U.S. Patent No. 11,238,344 - Artificially intelligent systems, devices, and methods for learning and/or using a device's circumstances for autonomous device operation

(Issued February 2, 2022)

The Invention Explained

  • Problem Addressed: The patent, which shares a specification with the ’974 Patent, addresses the same problem of prior systems being unable to learn from experience to achieve autonomous operation (’344 Patent, col. 1:39-47; Compl. ¶41).
  • The Patented Solution: Similar to the ’974 Patent, the invention describes a system that uses a knowledgebase of learned correlations between circumstances and instructions. An autonomous vehicle with sensors detects objects, and if a match is found with a previously learned circumstance representation, the vehicle performs the associated learned instructions, such as braking (’344 Patent, col. 163:36-164:5; Compl. ¶46).
  • Technical Importance: This technology allows for the creation of a shared knowledgebase that can be distributed across a fleet of vehicles, enabling one vehicle to benefit from the learned experiences of another (Compl. ¶49).

Key Claims at a Glance

  • The complaint asserts independent claim 1 and dependent claim 3 (Compl. ¶102).
  • Independent Claim 1 of the ’344 Patent recites essential elements including:
    • A memory storing a knowledgebase with a "first correlation" between a "first circumstance representation" and an instruction set.
    • The correlation is learned by operating a "first device" at least partially by a user.
    • The system receives a "second circumstance representation" from a "server" that receives it from a "second device."
    • A "third device" autonomously performs operations defined by the instruction set based on a partial match between the second and first circumstance representations.

Multi-Patent Capsules

  • U.S. Patent No. 11,055,583

    • Patent Identification: U.S. Patent No. 11,055,583, Machine learning for computing enabled systems and/or devices, issued July 6, 2021 (Compl. ¶55).
    • Technology Synopsis: The patent addresses systems overly dependent on user input. It discloses a machine learning solution where instruction sets are correlated with digital pictures. A device can then operate autonomously by capturing a new digital picture, matching it to a picture in the knowledgebase, and executing the corresponding instructions (Compl. ¶¶ 56, 58).
    • Asserted Claims: Independent claim 4 (Compl. ¶116).
    • Accused Features: Tesla’s Autopilot and Full Self-Driving (FSD) systems, which are alleged to use camera-based "Tesla Vision" and neural net processing (Compl. ¶¶ 32, 83).

  • U.S. Patent No. 10,102,449

    • Patent Identification: U.S. Patent No. 10,102,449, Devices, systems, and methods for use in automation, issued October 16, 2018 (Compl. ¶54).
    • Technology Synopsis: Similar to the ’583 Patent, this invention is directed to an artificial intelligence unit that correlates instruction sets with digital pictures. The AI unit receives a new digital picture, matches it against the stored, correlated pictures, and causes a device to autonomously execute the associated instruction set (Compl. ¶59).
    • Asserted Claims: Independent claims 1 and 17 (Compl. ¶109).
    • Accused Features: Tesla’s Autopilot and FSD systems in its vehicle fleet (Compl. ¶83).

  • U.S. Patent No. 10,607,134

    • Patent Identification: U.S. Patent No. 10,607,134, Artificially intelligent systems, devices, and methods for learning and/or using an avatar's circumstances for autonomous avatar operation, issued March 31, 2020 (Compl. ¶69).
    • Technology Synopsis: This patent extends the learning concept to simulated environments. It describes a system that learns correlations between object representations and instructions for a first "avatar" (e.g., a simulated vehicle). A second avatar can then operate autonomously within the simulation by detecting matching object representations and executing the learned instructions (Compl. ¶¶ 71, 73).
    • Asserted Claims: Independent claim 1 (Compl. ¶122).
    • Accused Features: Tesla's "Dojo" supercomputer and its associated simulation software used for AI training (Compl. ¶84).

  • U.S. Patent No. 11,113,585

    • Patent Identification: U.S. Patent No. 11,113,585, Artificially intelligent systems, devices, and methods for learning and/or using visual surrounding for autonomous object operation, issued September 7, 2021 (Compl. ¶70).
    • Technology Synopsis: This patent also addresses learning in a simulation but focuses on correlating instruction sets with digital pictures rather than object representations. A simulated object captures or renders digital pictures of its virtual surroundings and performs autonomous operations by matching those pictures to a learned knowledgebase of picture-instruction correlations (Compl. ¶¶ 72, 74).
    • Asserted Claims: Independent claim 1 (Compl. ¶129).
    • Accused Features: Tesla's "Dojo" supercomputer, which allegedly uses simulations based on real-world pictures and recreated synthetic worlds to train its AI (Compl. ¶¶ 84-85).

III. The Accused Instrumentality

  • Product Identification: The complaint identifies two categories of accused instrumentalities:
    1. Tesla Fleet Vehicles: Tesla Models S, 3, X, and Y equipped with "Software Version 9.0 and beyond," which includes Autopilot, Enhanced Autopilot, and Full Self-Driving (FSD) features (Compl. ¶83).
    2. Dojo Supercomputer: Tesla’s artificial intelligence training computer, referred to as "Dojo," and its associated simulation software (Compl. ¶84).
  • Functionality and Market Context:
    • The accused vehicles are equipped with cameras and a computer that uses neural net processing to analyze the vehicle's surroundings and perform autonomous driving functions (Compl. ¶37). A visual from Tesla's website, included in the complaint, illustrates the system's 360-degree camera visibility and sensor range (Compl. p. 34). These features are alleged to be of significant financial benefit to Tesla (Compl. ¶34).
    • The Dojo supercomputer is described as a "super powerful training computer" used to process vast amounts of video and other data collected from the Tesla vehicle fleet (Compl. ¶85). This data is used to train the AI for the autonomous driving systems. The complaint alleges Dojo is used to create simulations for training purposes, showing a "Scenario Reconstruction" pipeline that converts real-world clips into synthetic environments for testing and training (Compl. ¶84). A screenshot from a Tesla AI Day presentation shows a three-step process of "Real World Clip," "Auto-Labeled Reconstruction," and "Recreated Synthetic World" (Compl. p. 35).

IV. Analysis of Infringement Allegations

The complaint references exemplary infringement charts in Exhibits I-N, but these exhibits were not provided with the filing (Compl. ¶86). The infringement theory is therefore summarized from the complaint’s narrative allegations.

For the ’974 and ’344 Patents, the complaint alleges that the Tesla Fleet Vehicles embody the claimed systems (Compl. ¶¶ 96, 103). The infringement theory suggests that the collective operation of Tesla's fleet constitutes the "learning process" required by the claims (Compl. ¶17). Data from vehicles being operated by users (the "first device") is allegedly used to create and refine a knowledgebase (i.e., the trained AI model). This knowledgebase is then used by another vehicle (the "second device"), which detects its current environmental "circumstances" via sensors, matches that data against the learned knowledge, and autonomously performs driving operations like steering or braking (Compl. ¶¶ 44-45). The server-related elements of the ’344 Patent are allegedly met by Tesla’s central computing infrastructure, including the Dojo supercomputer, which processes fleet data and distributes updated models to the vehicles (Compl. ¶¶ 17, 85).

  • Identified Points of Contention:
    • Scope Questions: A central dispute may arise over the meaning of "circumstance representation." The patents provide examples of structured object data (e.g., object type, distance, bearing), raising the question of whether this term can be construed to read on the high-dimensional vector data processed by Tesla's neural networks.
    • Technical Questions: The claims recite a "second device" performing operations based on instruction sets learned from a "first device." This raises the question of whether the complaint provides evidence of a direct mapping between a specific learned experience from one car and a later autonomous action by another, or if Tesla's system relies on a generalized model trained on aggregated fleet data that does not preserve such a one-to-one correlation.

V. Key Claim Terms for Construction

  • The Term: "circumstance representation" (from claim 1 of the ’974 Patent).

  • Context and Importance: This term defines the data input to the claimed system. Its construction will be critical, as the infringement analysis depends on whether the sensor data processed by Tesla's Autopilot system falls within the scope of this term.

  • Intrinsic Evidence for Interpretation:

    • Evidence for a Broader Interpretation: The patent specification states that a circumstance representation may include "one or more object representations, or one or more collections of object representations" (’974 Patent, col. 172:12-15) and can be part of "a knowledge of how the device operated in a circumstance" (’974 Patent, col. 6:67-7:2). This language suggests a potentially broad scope beyond a simple list of objects.
    • Evidence for a Narrower Interpretation: The patent's primary embodiment, illustrated in Figure 4A, depicts a "Collection of Object Representations" as a table with discrete fields for "Type," "Distance," and "Bearing" for specific objects like a "Cat" or a "Person" (’974 Patent, Fig. 4A-4B; col. 80:19-67). This may support a narrower construction limited to structured, object-level data rather than the raw pixel or feature-vector data common in modern AI systems.

  • The Term: "a learning process that includes operating the first device at least partially by a user" (from claim 1 of the ’974 Patent).

  • Context and Importance: This term defines the mechanism for creating the knowledgebase. The parties may dispute whether Tesla's method of data collection from its fleet, which includes "shadow mode" operation and analysis of driver disengagements, constitutes the "learning process" envisioned by the patent.

  • Intrinsic Evidence for Interpretation:

    • Evidence for a Broader Interpretation: The patent background describes prior art systems as relying on a user to "direct their behaviors," suggesting that any user operation from which the system learns could satisfy this limitation (’974 Patent, col. 1:30-31). The complaint's theory of fleet learning, where the system learns from the collective actions of many drivers, may align with a broader interpretation (Compl. ¶17).
    • Evidence for a Narrower Interpretation: The claim language could be interpreted to require a more active or direct form of user-guided training, where user inputs are explicitly correlated with specific circumstances to generate instruction sets. This could be contrasted with Tesla's alleged large-scale, passive data aggregation from its entire fleet.

VI. Other Allegations

  • Indirect Infringement: The complaint alleges active inducement, stating that Tesla encourages customers to use the infringing Autopilot and FSD features through its marketing, customer presentations like "AI Day," user manuals, the Tesla mobile application, and in-vehicle displays (Compl. ¶¶ 87-90, 98).
  • Willful Infringement: The complaint alleges that Tesla has had knowledge of the asserted patents "since no later the filing of this Original Complaint," framing willfulness as a post-filing issue (Compl. ¶97). No facts are alleged to support pre-suit knowledge.

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

  • A core issue will be one of definitional scope: can the term "circumstance representation," which the patents illustrate using structured lists of discrete objects, be construed to cover the complex, high-dimensional vector data processed by the neural networks in Tesla's modern autonomous driving systems?
  • A second key question will concern architectural equivalence: does Tesla’s fleet-learning model—in which data from millions of vehicles is aggregated and trained on a central supercomputer to create a generalized AI—map onto the patent claims’ structure, which describes a specific correlation learned from a "first device" being used to command a "second device"?
  • For the patents directed to simulation, a central question will be one of functional mapping: does Tesla’s process of creating "recreated synthetic worlds" from real-world video clips for AI training on its Dojo supercomputer constitute the claimed methods of operating an "avatar" or "object" within a simulated application program?