IP Library Granted Patent US 12,632,512
Granted Patent B2
US 12,632,512 · App. 17/303,992 · Granted May 19, 2026

Ultrasonic system and method for tuning a machine learning classifier used within a machine learning algorithm

Inventors: Abinaya Kumar (Heimsheim, DE); Fabio Cecchi (Menlo Park, CA); Ravi Kumar Satzoda (Milpitas, CA); Lisa Marion Garcia (Santa Clara, CA); Mark Wilson (Pullman, WA); Naveen Ramakrishnan (Campbell, CA); Timo Pfrommer (Stuttgart, DE); Jayanta Kumar Dutta (Sunnyvale, CA); Juergen Johannes Schmidt (Magstadt, DE); Tobias Wingert (Leonberg, DE); Michael Tchorzewski (Böblingen, DE); Michael Schumann (Stuttgart, DE)
Assignee: Robert Bosch GmbH
G06F18/217G01S15/08G01S15/931G06F18/241G06N20/00
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,632,512
App. No.
17/303,992
Granted
May 19, 2026
Kind
B2
Abstract

A method and system is disclosed for tuning a machine learning classifier. An object class requirement may be provided and include rank thresholds. The object class requirements may also include a range goal that defines a minimum distance from the object the machine learning algorithm should not provide false positive results. A base classifier may be trained using a weighted loss function that includes one or more weight values that are computed using the one or more object class requirements. An output of the weighted loss function may be evaluated using an objective function which may be established using the one or more object class requirements. The one or more weights may also be re-tuned using the weighted loss function if the output of the weighted loss function does not converge within a predetermined loss threshold.

Claims (39)

1 . A method for tuning a machine learning classifier, comprising:

receiving one or more object class requirements, wherein the one or more object class requirements include one or more rank thresholds used to rank an object, and wherein the one or more object class requirements include (i) a traversability rank of the object and (ii) a range goal that defines a minimum distance from the object a machine learning algorithm should not provide false positive results;

training a base classifier using a weighted loss function, wherein the weighted loss function includes one or more weight values that are computed using the one or more object class requirements;

evaluating an output of the weighted loss function using an objective function, wherein the objective function is established with the one or more object class requirements;

tuning the one or more weight values using the weighted loss function if the output of the weighted loss function does not converge within a predetermined loss threshold;

maneuvering a vehicle to avoid the object in response to (i) the traversability rank of the object being less than a rank threshold and (ii) a distance between the vehicle and the object being less than the minimum distance; and

forgoing maneuvering the vehicle to avoid the object in response to (i) the traversability rank of the object being less than, greater than or equal to the rank threshold and (ii) the distance between the vehicle and the object being greater than the minimum distance.

2 . The method of claim 1 , wherein the weighted loss function is a cross entropy classification loss function.

3 . The method of claim 1 , wherein the object class requirements are established for an ultra-sonic sensor system.

4 . The method of claim 1 , wherein the object class requirements are established for differing vehicle platforms.

5 . The method of claim 1 , wherein the base classifier is used to determine a true positive output used for classifying the object.

6 . The method of claim 1 , wherein the base classifier is used to determine the false positive results used for classifying the object.

7 . The method of claim 1 , wherein the object class requirements are used to define the object having a specific geometry.

8 . The method of claim 1 , wherein the object class requirements are used to define the rank thresholds for the object having a rigid structure.

9 . The method of claim 1 , wherein the objective function is a squared distance to object error.

10 . The method of claim 1 , wherein the base classifier is used to define a predetermined distance between a vehicle and the object before a vehicle safety system is activated.

11 . The method of claim 1 further comprising forgoing maneuvering the vehicle to avoid the object such that the vehicle traverses the object in response to (i) the traversability rank of the object being greater than the rank threshold and (ii) the distance between the vehicle and the object being less than the minimum distance.

12 . A system for tuning a machine learning classifier, comprising:

one or more controllers configured to:

receive one or more object class requirements, wherein the one or more object class requirements include one or more rank thresholds used to rank an object classified by a machine learning algorithm, and wherein the one or more object class requirements include (i) a traversability rank of the object and (ii) a range goal that defines a minimum distance from the object the machine learning algorithm should not provide false positive results;

train a base classifier using a weighted loss function, wherein the weighted loss function includes one or more weight values that are computed using the one or more object class requirements;

evaluate an output of the weighted loss function using an objective function, wherein the objective function is established with the one or more object class requirements;

tune the one or more weight values using the weighted loss function if the output of the weighted loss function does not converge within a predetermined loss threshold,

maneuver a vehicle to avoid the object in response to (i) the traversability rank of the object being less than a rank threshold and (ii) a distance between the vehicle and the object being less than the mi m distance; and

forgo maneuvering the vehicle to avoid the object in response to (i) the traversability rank of the object being less than, greater than, or equal to the rank threshold and (ii) the distance between the vehicle and the object being greater than the minimum distance.

13 . The system of claim 12 , wherein the weighted loss function is a cross entropy classification loss function.

14 . The system of claim 12 , wherein the object class requirements are established for an ultra-sonic sensor system.

15 . The system of claim 12 , wherein the object class requirements are established for differing vehicle platforms.

16 . The system of claim 12 , wherein the base classifier is used to determine a true positive output used for classifying the object.

17 . The system of claim 12 , wherein the base classifier is used to determine the false positive results used for classifying the object.

18 . The system of claim 12 , wherein the controller is further configured to forgo maneuvering the vehicle to avoid the object such that the vehicle traverses the object in response to (i) the traversability rank of the object being greater than the rank threshold and (ii) the distance between the vehicle and the object being less than the minimum distance.

19 . A method for tuning a machine learning classifier, comprising:

receiving one or more object class requirements, wherein the one or more object class requirements include one or more rank thresholds used to rank an object classified by a machine learning algorithm based on received ultra-sonic sensor data used within a vehicle, and wherein the one or more object class requirements include (i) a traversability rank of the object and (ii) a range goal that defines a minimum distance from the object the machine learning algorithm should not provide false positive results:

training a base classifier using a weighted loss function, wherein the weighted loss function includes one or more weight values that are computed using the one or more object class requirements;

evaluating an output of the weighted loss function using an objective function, wherein the objective function is established with the one or more object class requirements;

tuning the one or more weight values using the weighted loss function if the output of the weighted loss function does not converge within a predetermined loss threshold; and

maneuvering the vehicle to avoid the object in response to (i) the traversability rank of the object being less than a rank threshold and (ii) a distance between the vehicle and the object being less than the minimum distance; and

forgoing maneuvering the vehicle to avoid the object in response to (i) the traversability rank of the object being less than, greater than, or equal to the rank threshold and (ii) the distance between the vehicle and the object being greater than the minimum distance.

20 . The method of claim 19 further comprising forgoing maneuvering the vehicle to avoid the object such that the vehicle traverses the object in response to (i) the traversability rank of the object being greater than the rank threshold and (ii) the distance between the vehicle and the object being less than the minimum distance.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jun 14, 2021
From: KUMAR, ABINAYA; CECCHI, FABIO; SATZODA, RAVI KUMAR; GARCIA, LISA MARION; WILSON, MARK; RAMAKRISHNAN, NAVEEN; PFROMMER, TIMO; DUTTA, JAYANTA; SCHMIDT, JUERGEN; WINGERT, TOBIAS; TCHORZEWSKI, MICHAEL; SCHUMANN, MICHAEL
To: ROBERT BOSCH GMBH
Reel/Frame 056527/0570 →
Continuity (1)
Related Publication 20220398414A1 · Dec 15, 2022
References Cited (6)
US 10891374B1 · Curtin et al. · 2021 [cited by applicant]
US 20140029840A1 · Jebara et al. · 2014 [cited by applicant]
US 20210046924A1 · Caldwell · 2021 [cited by examiner]
R. Theagarajan, F. Pala and B. Bhanu, “EDeN: Ensemble of Deep Networks for Vehicle Classification,” 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), Honolulu, HI, USA, 2017, pp. 906-913… [cited by examiner]
P. Saisan and S. Kadambe, “Shape normalized subspace analysis for underwater mine detection, ” 2008 15th IEEE International Conference on Image Processing, San Diego, CA, USA, 2008, pp. 1892-1895, doi: 10.1109/ICIP.2008… [cited by examiner]
Ibrahim, H.A., Azar, A.T., Ibrahim, Z.F., Ammar, H.H. (2020). “A Hybrid Deep Learning Based Autonomous Vehicle Navigation and Obstacles Avoidance.” Proceedings of the International Conference on Artificial Intelligence … [cited by examiner]