Patent Drafting Analysis of Tesla’s Detected Object Path Prediction for Vision-Based Systems | WO 2023/023336 A1

IP Drafting Analysis · WO 2023/023336 A1

Patent Drafting Analysis of Tesla's Detected Object Path Prediction for Vision-Based Systems | WO 2023/023336 A1

Q: What does WO 2023/023336 A1 protect?

WO 2023/023336 A1 protects a system and methods for predicting paths of travel for dynamic objects detected by vision-only vehicle systems, without relying on radar or LIDAR. The core innovation is obtaining separate ground truth label data for travel surface attributes (road edges, lane lines, road markings) and for detected dynamic object attributes, then processing these to generate a plurality of predicted paths each associated with a probabilistic confidence value. The claims further cover refining those predictions based on additional ground truth labels and storing the results for use by navigation or autonomous driving systems.

Q: Who owns patent WO 2023/023336 A1?

WO 2023/023336 A1 is owned by TESLA, INC., located at 1 Tesla Road, Austin, Texas 78725, United States. The inventors are Nanda Kishore Vasudevan, of 1 Tesla Road, Austin, Texas 78725 (US), and Dhiral Chheda, also of 1 Tesla Road, Austin, Texas 78725 (US).

Q: What are the independent claims of WO 2023/023336 A1?

Claim 1 is a system claim covering one or more computing systems implementing a vision system processing component that obtains separate travel-surface and dynamic-object ground truth label data from vision systems, processes them into a plurality of predicted paths with confidence values, further refines those paths based on an additional ground truth label, and stores the results. Claim 10 is a method claim with substantially the same steps as Claim 1 but without the additional-ground-truth-label processing step as a mandatory element of the independent claim. Claim 18 is a simplified method claim that combines travel surface and dynamic object ground truth into a single obtaining step, then generates and stores the plurality of predicted paths with confidence values.

Q: What technology does this patent cover?

This patent covers technology that allows a car equipped only with cameras — no radar or LIDAR sensors — to predict where nearby moving objects such as other vehicles or pedestrians are likely to travel next. The system analyzes camera images using machine learning to understand road markings, lane boundaries, and the speed and direction of detected objects, then generates multiple possible future paths for each detected object along with a probability score for each path. These predictions can be used to help the vehicle's navigation or self-driving system react appropriately, for example by anticipating that a car ahead might turn or change lanes.

Q: What IPC classifications does this patent use?

G06V 20/58 (2022.01) — Scenes, e.g. recognising scene from images; Semantic scene interpretation; Scene graphs; Identifying and classifying objects moving in three-dimensional scenes. G06V 20/56 (2022.01) — Context or environment of the image, e.g. vehicle occupants or pedestrians; Face and body posture; Objects not within field of view.

A structural and strategic analysis of Tesla's vision-only path prediction patent, examining claim architecture, drafting quality, critical gaps, and prosecution positioning across 21 claims spanning system and method formats.

WO 2023/023336 A1Filed: Aug 19, 2022Published: Feb 23, 2023G06V 20/58G06V 20/56

Draft a Patent Like This Explore Related Patents

Spec Words

5,820

Across 5 sections

Draft now ↗

Total Claims

3 independent · 18 dependent

Draft now ↗

Figure Sheets

Vehicle system, camera layout, processing architecture, path generation flow, path scenarios, feasibility cones

Draft now ↗

Published by PatSnap Insights Team · June 3, 2026 · 12 min read Verified by PatSnap Eureka Data

Overview

Structural Overview

The detailed description dominates at approximately 53% of total estimated words (~3,100 of ~5,820), providing moderate but not exhaustive technical depth across paragraphs [0011]–[0050]. The claim set comprises 21 claims across 3 independent claims — one system (Claim 1) and two method claims (Claims 10 and 18) — yielding an 18:3 dependent-to-independent ratio that is above the G06V software/AI norm. The 8 figure sheets (FIGs. 1A, 1B, 2, 3, 4A, 4B, 5A, 5B) provide coverage of system architecture, camera placement, processing components, the path generation routine, path scenarios with confidence values, and feasibility cone models.

Section Word Distribution

↗ Click bars to explore

Figure Inventory — 8 Sheets

Figure	Description	Role
FIG. 1A	High-level block diagram of vehicle 200 showing processing component(s) 214 interconnected with navigation, location, communication (COM), operational sensors, control components, and data store.Search in Eureka ↗	System architecture
FIG. 1B	Top-down vehicle diagram showing placement of front-facing cameras 202, door-pillar cameras 204, front bumper cameras 206, rear camera 208, and processing component(s) 212 inside vehicle 200.Search in Eureka ↗	Key embodiment
FIG. 2	Architecture diagram of processing component 212 showing processing unit 202, network interface 204, computer readable medium drive 206, I/O device interface 208, and memory 210 containing interface software 212, operating system 214, sensor interface component 216, path determination component 218, and navigational/driving component 220.Search in Eureka ↗	System architecture
FIG. 3	Flow diagram of predicted path generation routine 300 showing steps 302–314: obtain vision/ground truth data, identify travel surface attributes, identify dynamic objects, generate potential paths, further process based on additional criteria, store/transmit paths and confidence values.Search in Eureka ↗	Flow diagram
FIG. 4A	Road-scene illustration showing detected dynamic object 400 at a first point in time with three predicted paths of travel 402, 404, 406, each with 45% confidence value, depicting a left turn, straight travel, and right travel scenario.Search in Eureka ↗	Claim support
FIG. 4B	Road-scene illustration showing the same detected dynamic object 400 at a second point in time, with path 402 eliminated after passing the left-turn entry, leaving paths 404 and 406 each still at 45% confidence.Search in Eureka ↗	Claim support
FIG. 5A	Feasibility cone embodiment 502 for dynamic object 500 showing a narrow, elongated cone shape representing high-velocity/acceleration scenario where lateral maneuvers are constrained.Search in Eureka ↗	Claim support
FIG. 5B	Feasibility cone embodiment 504 for dynamic object 500 showing a wide, shorter cone shape representing low-velocity/acceleration scenario where more lateral maneuvers are feasible.Search in Eureka ↗	Claim support

Analysis powered by PatSnap Eureka. Patent text and figures publicly available from USPTO. Draft a Similar Patent

Claims

Claim Architecture Analysis

The claim set contains 3 independent claims: Claim 1 (system), Claim 10 (method), and Claim 18 (method), yielding two independent method claims and one system claim but notably no computer-readable medium (CRM) claim. The dependent-to-independent ratio of 6:1 is above the typical G06V software/AI norm of 4–5:1, providing reasonable fallback depth. The dual method claim strategy (Claims 10 and 18) is strategically unusual — Claim 18 is a simplified, compressed variant of Claim 10 that combines travel surface and dynamic object ground truth into a single step, suggesting an intentional ladder approach but creating redundancy rather than meaningful differentiation.

Core inventive concept: The claims address the problem of predicting how detected dynamic objects will move relative to a vision-only vehicle system — without radar or LIDAR — by processing separately obtained first ground truth label data (travel surface attributes: road edges, lane lines, road markings) and second ground truth label data (dynamic object attributes) from collected vision data to generate a plurality of predicted paths of travel each associated with a confidence value, then further refining those paths based on at least one additional ground truth label. The confidence value architecture — including the expressly claimed scenario where the sum of confidence values can exceed 100% (Claim 7) — is a specific technical differentiator recited in Claim 1 and elaborated in Claims 7, 8, and 9.

Independent Claim Dissection

Claim	Preamble	Transition	Key Body Elements
Claim 1	A system for managing vision systems in vehicles, the system	comprising:	one or more computing systems with processing devices and memory executing a vision system processing component; obtain first ground truth label data (travel surface attributes including road edges, lane lines, road markings); obtain second ground truth label data (dynamic object attributes); process first and second ground truth label data to form a plurality of predicted paths each with confidence value; process plurality based on at least one additional ground truth label; store the processed plurality of predicted paths and associated confidence valuesSearch prior art ↗
Claim 10	A method for managing vision systems in vehicles, the system	comprising:	obtaining first ground truth label data (travel surface attributes); obtaining second ground truth label data (dynamic object attributes); processing first and second ground truth label data to form a plurality of predicted paths each with confidence value; storing the processed plurality of predicted paths and associated confidence valuesSearch prior art ↗
Claim 18	A method for managing vision systems in vehicles, the system	comprising:	obtaining ground truth label data (both travel surface attributes and dynamic object attributes combined); generating a plurality of predicted paths based on obtained ground truth label data each with confidence value; storing the processed plurality of predicted paths and associated confidence valuesSearch prior art ↗

Claim Dependency Tree

1 System — computing system with vision processing component; obtains first/second ground truth labels; generates and stores plurality of predicted paths with confidence valuesSearch Claim 1 prior art ↗

2 Adds: selecting paths exceeding a minimal confidence value thresholdSearch in Eureka ↗

3 Adds: first and second ground truth label data corresponds to objects detected within a horizon of captured video dataSearch in Eureka ↗

4 Further: (depends on 3) objects detected beyond current defined location of vehicleSearch in Eureka ↗

5 Adds: dynamic object attributes correspond to at least one of yaw, velocity, or accelerationSearch in Eureka ↗

6 Adds: further processing by identifying at least one static object that may interfere with a predicted pathSearch in Eureka ↗

7 Adds: sum of confidence values for two or more predicted paths exceeds 100%Search in Eureka ↗

8 Adds: sum of confidence values for the plurality does not exceed 100%Search in Eureka ↗

9 Adds: processing plurality based on modeled feasibility cone for a detected dynamic objectSearch in Eureka ↗

10 Method — obtaining first/second ground truth labels; processing to form plurality of predicted paths with confidence values; storing resultsSearch Claim 10 prior art ↗

11 Adds: forming paths by selecting those exceeding minimal confidence thresholdSearch in Eureka ↗

12 Adds: ground truth label data corresponds to objects within horizon of captured video dataSearch in Eureka ↗

13 Adds: first ground truth label data corresponds to road edges, lane lines, or road markingsSearch in Eureka ↗

14 Adds: dynamic object attributes correspond to at least one of yaw, velocity, or accelerationSearch in Eureka ↗

15 Adds: further comprising processing the plurality based on at least one additional ground truth labelSearch in Eureka ↗

16 Further: (depends on 15) further processing includes identifying at least one static object that may interfere with a predicted pathSearch in Eureka ↗

17 Adds: processing plurality based on modeled feasibility cone for detected dynamic objectSearch in Eureka ↗

18 Method (simplified) — obtaining combined ground truth label data (travel surfaces and dynamic objects); generating plurality of predicted paths with confidence values; storing resultsSearch Claim 18 prior art ↗

19 Adds: forming paths by selecting those exceeding minimal confidence thresholdSearch in Eureka ↗

20 Adds: first ground truth label data corresponds to road edges, lane lines, or road markingsSearch in Eureka ↗

21 Adds: dynamic object attributes correspond to at least one of yaw, velocity, or accelerationSearch in Eureka ↗

Metric	This Application	Software / AI / Autonomous Vehicle Norm
Total claims	21	15 – 25
Independent claim count	3	2 – 4
Dependent : Independent ratio	6.00 : 1	4 – 8 : 1
Method claims present?	Yes — Claims 10, 18	Common
System / apparatus claims?	Yes — Claim 1	Common

Analysis powered by PatSnap Eureka. Patent text and figures publicly available from USPTO. Draft a Similar Patent

Drafting Quality

Drafting Quality Signals

The claim set demonstrates clear structural logic in separating travel-surface ground truth (first label data) from dynamic-object ground truth (second label data) across Claims 1 and 10, providing a technically defensible distinction over prior art that processes fused sensor data. However, the complete absence of a computer-readable medium (CRM) claim and the redundancy between Claims 10 and 18 — which share nearly identical core limitations — represent material prosecution vulnerabilities that could narrow Tesla's enforcement options.

✅

Antecedent Basis

The claim language is generally clean with respect to antecedent basis. In Claim 1, "the obtained first ground truth label data" and "the obtained second ground truth label data" correctly refer back to the elements introduced by "obtain first ground truth label data" and "obtain second ground truth label data." Similarly, "the process plurality of predicted paths" in the store step refers back to the earlier "form a plurality of predicted paths of travel" element. No orphaned "the" references were identified across Claims 1–21, which reduces the risk of indefiniteness rejections under 35 U.S.C. §112(b).

Explore in Eureka ↗

✅

Spec–Claim Consistency

Key Claim 1 limitations map directly to specific specification paragraphs and figures. The "obtain first ground truth label data" limitation (road edges, lane lines, road markings) is supported by ¶[0037] and FIG. 3 step 304. The "obtain second ground truth label data" (dynamic objects) is supported by ¶[0038] and FIG. 3 step 306. The confidence value limitation is supported by ¶[0041] and illustrated in FIGs. 4A/4B. The feasibility cone limitation of Claim 9 is explicitly supported by ¶[0040] and FIGs. 5A/5B. The mapping is consistent, reducing §112(a) written description risk for all independent claim limitations.

Explore in Eureka ↗

✅

Transition Word Usage

All independent claims (Claims 1, 10, 18) use "comprising" as the transition, which is the strategically optimal choice for this technology space as it permits the claimed system or method to include additional components or steps without negating infringement. The specification at ¶[0047] further reinforces this by explicitly noting that "comprising", "including", and "incorporating" are intended to be construed in a non-exclusive manner. No missed opportunities for "consisting essentially of" or narrower transitions were identified that would have been preferable — the open-ended transitions correctly preserve scope for downstream infringement assertions against implementers who add supplementary sensing modules.

Explore in Eureka ↗

✅

§112(f) Means-Plus-Function Risk

No "means for" or "step for" language appears anywhere in Claims 1–21, eliminating direct §112(f) exposure. The functional language used — "operative to: obtain", "processing", "storing" — is tied to the structural antecedent "one or more computing systems including processing devices and memory" in Claim 1, which provides adequate structural grounding under post-Williamson v. Citrix jurisprudence. The term "vision system processing component" in Claim 1 could attract scrutiny as a nonce word, but the specification at ¶[0034] defines this component as software residing in memory 210 (path determination component 218), providing the structural disclosure needed to rebut a §112(f) challenge.

Explore in Eureka ↗

⚠️

§101 Eligibility Risk

This patent faces material Alice/Mayo §101 exposure because Claims 10 and 18 are pure method claims directed to data processing steps — obtaining, processing, and storing ground truth label data — with no hardware tie-in beyond generic computing systems. While Claim 1 anchors the system to "one or more computing systems including processing devices and memory," this level of structural specificity has been found insufficient by the USPTO in similar autonomous vehicle perception claims. The claimed confidence value generation and feasibility cone modeling (Claims 9, 17) are the strongest candidates for an "inventive concept" defense under Alice Step 2B, but neither is recited in the broadest independent claims, creating a significant prosecution vulnerability for Claims 10 and 18 as filed.

Explore in Eureka ↗

⚠️

Dependent Claim Fallback Quality

The dependent claims provide some meaningful fallback positions — notably Claim 9 (feasibility cone modeling), Claims 7/8 (confidence value sum behavior), and Claim 6 (static object interference detection) — but are significantly weakened by triplication: Claims 5, 14, and 21 are substantively identical (yaw/velocity/acceleration attributes), and Claims 2, 11, and 19 mirror each other (confidence threshold filtering), and Claims 6 and 16 share nearly the same limitation (static object interference). This pattern adds word count but does not create distinct fallback positions; it reflects mechanical parallel drafting across the three independent claims rather than strategic claim architecture that would provide genuine coverage differentiation if any independent claim is invalidated.

Explore in Eureka ↗

⚠️

Abstract Quality

An examiner reading only the abstract would identify the general concept (vision-based path prediction with confidence values) but would fail to identify the novel contribution — the two-step ground truth label acquisition architecture (separate travel surface labels and dynamic object labels) that structurally differentiates Claim 1 from prior art. The abstract states that "a service can process the set of inputs (e.g., the associated ground truth label data)" but does not distinguish first from second ground truth label data, omits the additional-label processing step that is core to Claim 1, and does not mention the feasibility cone or the non-summing-to-100% confidence value architecture that appear in the most technically distinctive dependent claims.

Explore in Eureka ↗

✅

Figure Support Quality

The 8 figure sheets provide solid coverage of the major structural claim limitations. FIG. 2 directly supports the computing system architecture of Claim 1 (processing unit 202, memory 210, path determination component 218, sensor interface component 216). FIG. 3 maps to the method steps of Claims 10 and 18 step-by-step (steps 302–312). FIGs. 4A/4B support the confidence value limitations and the dynamic path update mechanism. FIGs. 5A/5B support the feasibility cone limitation of Claim 9. The primary gap is that no figure explicitly illustrates the separation of first ground truth label data (travel surfaces) from second ground truth label data (dynamic objects), which is the most critical structural distinction of Claim 1 over the cited prior art (US 2019/096256).

Explore in Eureka ↗

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

Spec–Claim Consistency

3.8

Dependent Claim Coverage

2.8

Claim Type Diversity

2.5

Figure Support Quality

3.5

Key observation: Spec–Claim Consistency scores highest (3.8/5) because FIGs. 2, 3, 4A/4B, and 5A/5B map directly to the major independent claim limitations via specific paragraphs [0034]–[0045], providing solid §112(a) written description support. Claim Type Diversity scores lowest (2.5/5) because the complete absence of a computer-readable medium (CRM) claim leaves a significant enforcement gap — any competitor implementing the path prediction algorithm in downloadable automotive software would not be reachable by the current claim set. Practitioners should consider filing a continuation that adds CRM claims and refactors the duplicate method claims (10 and 18) into a single method claim with broader scope.

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.

GAP 01 · HIGHEST IMPACT

No computer-readable medium claim leaves software distribution uncovered

The claim set contains one system claim (Claim 1) and two method claims (Claims 10, 18) but no computer-readable medium (CRM) or non-transitory storage medium claim, creating a complete enforcement blind spot for software-based implementations. Any autonomous driving stack provider that licenses or distributes Tesla's path prediction algorithm as downloadable firmware or OTA update packages would not be reachable by the current claims because neither a system claim (requires a deployed computing system) nor a method claim (requires performance of the steps) would attach to the act of distributing the stored instructions. A stronger filing would have included a Beauregard-style CRM claim mirroring Claim 1's operative steps as instructions stored on non-transitory computer-readable media, which is standard practice in automotive AI patents and directly supported by FIG. 2's depiction of computer readable medium drive 206 and memory 210.

GAP 02 · HIGH IMPACT

Vision-only restriction absent from independent claims enables easy design-around

Claims 1, 10, and 18 do not affirmatively require that the vision system operate without radar or LIDAR — despite the specification at ¶[0011] and ¶[0019] explicitly defining the invention as a "vision-only" system contrasted against radar/LIDAR-based systems. A competitor could implement an equivalent path prediction system using vision plus supplementary LIDAR and argue non-infringement on the basis that their system is not limited to vision-only inputs, even though all the claimed processing steps are identical. The ISR cited prior art (WO 2020/245654, Mobileye) that uses multi-sensor fusion for path prediction — a stronger filing would have positively recited "wherein the one or more vision systems are the sole sensor input" or "without any radar or LIDAR detection systems" in Claim 1 to capture the technical contribution and foreclose the multi-sensor design-around.

GAP 03 · HIGH IMPACT

Redundant method claims 10 and 18 waste claim scope without differentiation

Unlock to read the full analysis.

🔒

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.

Missing CRM claim for software distribution Vision-only restriction absent from claims Redundant method claims waste scope

Unlock Full Analysis — Free

Frequently asked questions

WO 2023/023336 A1 — key questions answered

What does WO 2023/023336 A1 protect?+

Who owns patent WO 2023/023336 A1?+

What are the independent claims of WO 2023/023336 A1?+

What technology does this patent cover?+

What IPC classifications does this patent use?+

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.

Link copied to clipboard

AI AGENTS

AI APPLICATIONS

OTHERS

INDUSTRIES

USE CASES

EXPLORE

ENGAGE

SUPPORT & SERVICES

Your Agentic AI Partner
for Smarter Innovation

Great, Please verify your email.

Patent Drafting Analysis of Tesla’s Detected Object Path Prediction for Vision-Based Systems | WO 2023/023336 A1

Patent Drafting Analysis of Tesla's Detected Object Path Prediction for Vision-Based Systems | WO 2023/023336 A1