Patent Drafting Analysis of Tesla’s Ground Truth Generation for Autonomous Driving ML | US 10,997,461 B2
Patent Drafting Analysis of Tesla's Ground Truth Generation for Autonomous Driving ML | US 10,997,461 B2
A structural and strategic analysis of Tesla's time-series-based ground truth generation patent, examining claim architecture, drafting quality, critical gaps, and prosecution positioning across method, CRM, and system claim types.
Spec Words
5,820
Across 5 sections
Draft now ↗
Total Claims
20
3 independent · 17 dependent
Draft now ↗
Figure Sheets
6
System block diagram, process flow diagrams, sensor images with lane line annotations
Draft now ↗
Published by PatSnap Insights Team · · 12 min read Verified by PatSnap Eureka Data
Overview
Structural Overview
The detailed description dominates at approximately 67% of total words (~3,900 of ~5,820), reflecting a thorough operational narrative but leaving the summary and background sections comparatively thin. The claim set comprises 20 claims — 3 independent (method, CRM, system) and 17 dependent — providing a tripartite structural coverage with a 5.67:1 dependent-to-independent ratio. Six drawing sheets cover the system architecture (FIG. 1), two process flow diagrams (FIGs. 2, 3, 4), and two annotated real-world sensor images (FIGs. 5, 6) that directly illustrate the lane line time-series ground truth concept central to the invention.
Section Word Distribution
↗ Click bars to exploreFigure Inventory — 6 Sheets
| Figure | Description | Role |
|---|---|---|
| FIG. 1 | Block diagram of deep learning system 100 for autonomous driving, showing sensors 101, image pre-processor 103, deep learning network 105, AI processor 107, vehicle control module 109, and network interface 111.Search in Eureka ↗ | System architecture |
| FIG. 2 | Flow diagram illustrating a six-step process for training and applying a machine learning model for autonomous driving, from preparing training data (201) through controlling the vehicle (211).Search in Eureka ↗ | Flow diagram |
| FIG. 3 | Flow diagram illustrating a four-step process for creating training data using a time series of elements, including receiving elements (301), receiving related data (303), determining ground truth (305), and packaging training data (307).Search in Eureka ↗ | Claim support |
| FIG. 4 | Flow diagram illustrating a six-step process for collecting sensor and odometry data on a vehicle and transmitting it for training data generation, from receiving sensor data (401) to transmitting sensor and related data (411).Search in Eureka ↗ | Flow diagram |
| FIG. 5 | Annotated real-world camera image 500 showing vehicle lane lines 501 and 511 with labeled portions 503/513, 505/515, 507/517 captured at time-series locations A, B, and C, illustrating how ground truth is assembled from multiple time-series frames.Search in Eureka ↗ | Key embodiment |
| FIG. 6 | Annotated real-world camera image 600 showing predicted three-dimensional trajectories of lane lines 601 and 611 (including distant portions 621) overlaid on sensor data, demonstrating the trained model's inference capability from a single input frame.Search in Eureka ↗ | Key embodiment |
Analysis powered by PatSnap Eureka. Patent text and figures publicly available from USPTO. Draft a Similar Patent
Claims
Claim Architecture Analysis
The patent contains 3 independent claims: Claim 1 (method), Claim 19 (computer program product/CRM), and Claim 20 (system), providing full tripartite enforcement coverage across process, software, and hardware embodiments. The 17 dependent claims yield a 5.67:1 dependent-to-independent ratio, which is above the norm for software/AI art units (typically 4–6:1), indicating reasonable fallback depth. The claim strategy focuses dependent claims almost entirely on Claim 1 (Claims 2–18), leaving Claims 19 and 20 as bare independent claims with no dependent refinements, which creates a significant defensive gap for the CRM and system branches.
Core inventive concept: The independent claims solve the problem of generating accurate machine learning training labels without costly manual annotation by determining a ground truth for a selected time series element based on the collective information from a plurality of time series elements — specifically by "identifying, for individual time series elements, respective portions of the individual time series elements to form the ground truth" — and then training a model to predict that group-derived ground truth from only a single input element.
Independent Claim Dissection
| Claim | Preamble | Transition | Key Body Elements |
|---|---|---|---|
| Claim 1 | A method, comprising | comprising | receiving sensor data including a group of time series elements captured at respective times within a period of time; determining a ground truth by identifying respective portions of individual time series elements and generating ground truth based on identified portions; using a processor to train a machine learning model using a training dataset comprising the ground truth and a selected single time series element, wherein model is trained to output ground truth based on input of selected elementSearch prior art ↗ |
| Claim 19 | A computer program product, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions which when executed by a processor, cause the processor to | comprising | receive sensor data including a group of time series elements; determine ground truth by identifying respective portions and generating ground truth; form a training dataset limited to determined ground truth and a single selected time series element; training dataset configured to train a machine learning model to output ground truth based on input of single time series elementSearch prior art ↗ |
| Claim 20 | A system, comprising | comprising | a processor; and a memory coupled with the processor configured to provide instructions which cause the processor to: receive sensor data including a group of time series elements; determine ground truth by identifying respective portions and generating ground truth; train a machine learning model using training dataset comprising determined ground truth and a selected time series element, wherein model trained to output ground truth based on input of selected elementSearch prior art ↗ |
Claim Dependency Tree
1 Method: receive time series sensor data, determine ground truth from group portions, train ML model on selected single elementSearch Claim 1 prior art ↗
2 Adds: ground truth associated with a vehicle lane lineSearch in Eureka ↗
3 Further: identifying portions comprises identifying different portions of the lane line from different time series elementsSearch in Eureka ↗
4 Adds: group of time series elements used to identify locations of a vehicle lane line in the selected elementSearch in Eureka ↗
5 Adds: each element includes an image associated with a corresponding timestampSearch in Eureka ↗
6 Adds: ground truth determined using odometry data associated with the groupSearch in Eureka ↗
7 Further: odometry data includes vehicle position data and vehicle orientation dataSearch in Eureka ↗
8 Further: odometry data identifies a first change in vehicle position and a second change in vehicle orientationSearch in Eureka ↗
9 Adds: ground truth represents a three-dimensional trajectory of a lane lineSearch in Eureka ↗
10 Further: three-dimensional trajectory represented as a parametric curveSearch in Eureka ↗
11 Adds: ground truth associated with a predicted path of a second vehicle different from the first vehicle with sensorsSearch in Eureka ↗
12 Further: second vehicle identified as entering a lane of the first vehicleSearch in Eureka ↗
13 Adds: ground truth associated with a distance of an objectSearch in Eureka ↗
14 Further: object is an obstacle, moving vehicle, stationary vehicle, or barrierSearch in Eureka ↗
15 Further: distance of object determined based on radar data associated with the groupSearch in Eureka ↗
16 Adds: number of elements in group based on distance traveledSearch in Eureka ↗
17 Adds: number of elements in group based on average vehicle speedSearch in Eureka ↗
18 Adds: identifying a portion of an individual element is based on an accuracy measure associated with a feature depicted in the ground truthSearch in Eureka ↗
19 CRM: non-transitory computer readable storage medium with instructions to receive time series sensor data, determine ground truth from portions, form training dataset limited to ground truth and single selected elementSearch Claim 19 prior art ↗
20 System: processor and memory configured to receive time series sensor data, determine ground truth from portions, train ML model using ground truth and selected elementSearch Claim 20 prior art ↗
| Metric | This Application | Software / AI / Autonomous Vehicle Norm |
|---|---|---|
| Total claims | 20 | 15 – 25 |
| Independent claim count | 3 | 2 – 4 |
| Dependent : Independent ratio | 5.67 : 1 | 4 – 7 : 1 |
| Method claims present? | Yes — Claim 1 | Always |
| System / apparatus claims? | Yes — Claim 20 | Common |
Analysis powered by PatSnap Eureka. Patent text and figures publicly available from USPTO. Draft a Similar Patent
Drafting Quality
Drafting Quality Signals
Tesla's prosecution strategy demonstrates notable strengths in the method claim's layered fallback structure (Claims 2–18 off Claim 1) and in the real-world figure support provided by FIGs. 5 and 6 for the core ground truth concept, but is weakened by the complete absence of dependent claims under the CRM Claim 19 and system Claim 20, leaving these independent claims unprotected by any fallback positions. The §101 eligibility exposure is partially mitigated by the autonomous vehicle control hardware tie-in, but the core training step in Claim 1 is framed in purely functional terms that an examiner could characterize as abstract.
✅
Antecedent Basis
Antecedent basis is clean throughout the claim set. In Claim 1, "a group of time series elements" is introduced in the receiving step, and all subsequent references use "the group of time series elements" and "the ground truth" with proper antecedents. Claim 19's "the determined ground truth" and "the selected time series element" both track cleanly to their introducing clauses. No orphaned "the" references were identified across all 20 claims.
✅
Spec–Claim Consistency
The specification provides strong support for the core Claim 1 limitations. FIG. 3 (steps 301–307) maps directly to the receive-determine-train sequence of Claim 1. FIG. 5 and its description (col. 15–17) explicitly support the "identifying respective portions of individual time series elements" limitation, showing portions 503/513, 505/515, and 507/517. The odometry data limitation of Claim 6 is supported by the detailed discussion at col. 11. The parametric curve limitation of Claim 10 is supported by col. 12 ("three-dimensional parameterized spline or curve").
✅
Transition Word Usage
All three independent claims correctly use "comprising" as the transitional phrase, preserving open-ended scope and allowing infringers who add additional steps or components to still be captured. Claim 19 is structured as a CRM with "comprising computer instructions" which is appropriate. No instances of "consisting of" or "consisting essentially of" appear, which is strategically correct for a software and systems patent where implementation details vary widely across potential infringers.
⚠️
§112(f) Means-Plus-Function Risk
A potential §112(f) concern arises in Claim 20's "memory configured to provide the processor with instructions which when executed cause the processor to" language — while this is standard system claim drafting, courts have occasionally scrutinized such functional claiming for adequate structural definition. More problematically, Claim 1's limitation "using a processor to train a machine learning model" describes only a functional result without structural specificity; if an examiner reads this as a means-plus-function limitation (no "means for" language but potentially invocable), the corresponding structure in the spec (AI processor 107, deep learning network 105) would need to be the defined corresponding structure, narrowing scope significantly.
⚠️
§101 Eligibility Risk
Claim 1 faces Alice step-two exposure because the core method — receiving sensor data, determining a ground truth from time series, and training a model — could be characterized as an abstract idea of "data collection and mathematical processing." The hardware tie-in is weak in Claim 1's preamble (no vehicle or sensor is recited in the preamble) and only the phrase "using a processor" appears in the training step. Claim 20's processor-and-memory structure provides stronger §101 footing, and the autonomous vehicle control context in the specification (though not in Claim 1) provides an inventive concept argument, but a stronger filing would have recited the vehicle-mounted sensor configuration directly in Claim 1.
⚠️
Dependent Claim Fallback Quality
All 17 dependent claims depend exclusively from Claim 1, leaving Claims 19 and 20 as bare independent claims without any dependent fallback — a significant prosecution and litigation weakness. Among the method claim dependents, Claims 9 and 10 (3D trajectory and parametric curve representation) add meaningful technical specificity tied to FIG. 6, and Claims 6–8 (odometry data) create a valuable secondary fallback. However, Claims 16 and 17 (number of elements based on distance or speed) are relatively thin limitations easily designed around, and the pair adds minimal prosecution value over each other.
⚠️
Abstract Quality
The abstract states that "a training data set is determined, including by determining for at least a selected time series element in the group of time series elements a corresponding ground truth" which accurately identifies the general concept but fails to name the specific technical mechanism — the multi-element portion identification and synthesis — that distinguishes this from prior art. An examiner reading only the abstract would likely characterize the invention as generic "machine learning training data preparation," missing the novel insight that ground truth is derived from the collective accuracy of a plurality of time series elements and then associated back to a single selected element.
✅
Figure Support Quality
Figure support is strong for the core claim limitations. FIG. 1 supports the system claim's processor/memory/sensor components (Claim 20). FIGs. 3 and 5 together provide the clearest visual support for the "identifying respective portions of individual time series elements" limitation in Claims 1 and 19. FIG. 6 directly supports Claims 9 and 10 (three-dimensional trajectory and parametric curve). The odometry-related Claims 6–8 are supported by the specification text but lack a dedicated figure showing odometry data integration — a minor gap that is compensated by the detailed written description at col. 11.
Analysis powered by PatSnap Eureka. Patent text and figures publicly available from USPTO. Draft a Similar Patent
Scorecard
Strategic Intent Scorecard
Multi-dimensional assessment of this application's patent strategy quality, based on claim structure, specification depth, and prosecution positioning.
Claim Breadth
3.5
Prosecution Defensibility
3
Spec–Claim Consistency
4.2
Dependent Claim Coverage
2.8
Claim Type Diversity
4
Figure Support Quality
3.8
Key observation: Spec–Claim Consistency scores highest (4.2/5.0) because the specification provides detailed, figure-anchored support for every major claim limitation — FIGs. 3, 5, and 6 directly illustrate the ground truth assembly mechanism of Claim 1, and the odometry limitations of Claims 6–8 are thoroughly described in the written description. Dependent Claim Coverage scores lowest (2.8/5.0) because all 17 dependent claims hang exclusively from method Claim 1, leaving the CRM (Claim 19) and system (Claim 20) independent claims entirely unprotected by any dependent fallback positions, which means that if Claims 1 and 20 are invalidated on different grounds, Tesla has no dependent claim safety net for the system and CRM branches. Practitioners should note that a continuation application adding dependent claims to Claims 19 and 20 would dramatically strengthen the portfolio's defensive depth.
See how your own draft compares — Open Eureka IP Drafting →
Critical Gaps
3 Critical Gaps in This Claim Set
A senior-attorney lens on the three highest-priority structural weaknesses — what each exposes in prosecution and litigation, and what a stronger filing would have done differently.
🔒
3 Critical Gaps in This Claim Set
See the full attorney-level analysis of what this application leaves unprotected — and how to draft it more defensively for your own filings.
No dependents under CRM and system claims Claim 1 lacks §101 hardware tie-in No claims on trigger-based training data capture
Unlock Full Analysis — Free Frequently asked questions
US 10,997,461 B2 — key questions answered
Still have questions? PatSnap Eureka can answer them from patent data instantly. Search in Eureka
Disclaimer: This analysis is generated by PatSnap Eureka AI based on publicly available patent data from the USPTO. It does not constitute legal advice and should not be relied upon as such. Patent data may be subject to change as prosecution progresses. Scores and assessments reflect automated analysis and may not capture all relevant legal or technical nuances. Always consult a qualified patent attorney for formal legal opinions on patentability, freedom to operate, or infringement.
Ask anything about this patent.
PatSnap Eureka searches patents and data to answer instantly.
PatSnap Eureka searches patents and data to answer instantly.
Powered by PatSnap Eureka
Link copied to clipboard