Title
Sonal Beladar
Guide -Dr. Sujit N. Deshpande
Seminar - I
PES Modern College of Engineering
Department of Computer Engineering
Savitribai Phule Pune University
dd-JAN 2025
Dept of Comp Engg Short Title dd-October-2024 1 / 15
Outline
1 Introduction
2 Literature survey
3 Objectives and Scop
4 e
5 Problem Statement
6 Architectural Diagram
7 Algorithmic Approach
8 Input / Data sets
9 Results/Experimental Wor
k
Conclusion / Summary
References
Dept of Comp Engg Short Title dd-October-2024 2 / 15
Introduction
In recent years, the field of Natural Language Processing (NLP) has experienced
significantadvancements, enabling machines to interact with humans in a more
natural and human-like manner.
Applications such as smart assistants, intelligent search engines, and automated
customer supportsystems have showcased the transformative potential of NLP.
Despite their varied contexts, thesesystems rely on core advancements in deep
learning and machine learning technologies, which havesignificantly improved their
performance and applicability.
The integration of AI in autonomous vehicles is revolutionizing transportation,
enabling safety,efficiency, and accessibility.
However, challenges such as ethical dilemmas, cyber security, and scalability
demand ongoing innovation.
Dept of Comp Engg Short Title dd-October-2024 3 / 15
Literature Survey
Table 1:-Literature Survey
Dept of Comp Engg Short Title dd-October-2024 4 / 15
Objectives and Scope
To develop a method for mining disengagement reports and extracting
meaningful information.
To create a tool that converts this information into test scenarios in the
Open Scenario format, a standard for AV simulation testing.
To evaluate the effectiveness of NLP techniques in handling large-scale
textual data and generating actionable scenarios.
Dept of Comp Engg Short Title dd-October-2024 5 / 15
Problem Statement
Examine the evolution of AI algorithms in AVs and their
increasing complexity Analyze the Operational Design
Domain (ODD) diversity for various vehicle types.
Highlight challenges such as safety, cyber security, and
ethical considerations in AI-driven systems.
Dept of Comp Engg Short Title dd-October-2024 6 / 15
Architectural Diagram
Figure: [Link] Architectural Diagram.
Dept of Comp Engg Short Title dd-October-2024 7 / 15
Algorithmic Approach(1/2)
BERT (Bidirectional Encoder Representations from
Transformers)
1. Initialize BERT model with pre-trained weights.
2. Tokenize input text into subwords.
3. Pass tokens through BERT layers to obtain contextual
embeddings.
4. Apply classification layer on embeddings.
5. Output predicted class probabilities.
Dept of Comp Engg Short Title dd-October-2024 8 / 15
Algorithmic approach(2/2)
Data Collection and Preprocessing
1. Collect disengagement reports, such as those from the California DMV,
containing details of
AV disengagements Tokenize input text
2. Preprocess the data to ensure quality and uniformity Apply a linear layer to
generate logits
3. Text Cleaning: Handle incomplete, noisy, or ambiguous data.
4. Tokenization: Split textual data into meaningful tokens for analysis.
5. Normalization: Standardize text formats to enable consistent processing.
NLP Techniques
1. Text Parsing: Extract structured information
(e.g., disengagement reasons, vehicle behaviors, environmental conditions). 2.
Pattern Matching: Identify recurring phrases and keywords to classify events
(e.g., "sensor malfunction" or "obstacle detection failure"). 3. Entity
Recognition: Detect critical entities such as vehicle types, road conditions, and
obstacles.
Dept of Comp Engg Short Title dd-October-2024 9 / 15
Future challenges
[Link] and Reliability: Ensuring flawless AI
performance in diverse conditions.
o Mitigation: Robust sensor fusion and redundancy.
2. Cybersecurity: Preventing hacking and unauthorized
access.
o Mitigation: Strong encryption, penetration testing, and
secure communication protocols.
3. Ethical Dilemmas: Addressing biases in AI models
and ensuring fair decision-making.
o Mitigation: Implementing explainable AI (XAI) and
diverse training datasets.
4. Public Trust: Building confidence through
transparency and rigorous testing.
Dept of Comp Engg Short Title dd-October-2024 10 /
Conclusion
● The integration of AI in autonomous vehicles is
revolutionizing transportation, enabling safety, efficiency, and
accessibility. However, challenges such as ethical dilemmas,
cybersecurity, and scalability demand ongoing innovation.
● The future of AVs lies in harnessing emerging AI technologies
to achieve higher levels of autonomy while addressing
societal and regulatory concernsThe study demonstrates the
potential of NLP in automating the generation of test
scenarios for AV testing.
Dept of Comp Engg Short Title dd-October-2024 11 /
References
1 Prashant Upadhyay ,Rishabh Agarwal,Sumeet Dhiman, Abhinav Sarkar ,Saumya
Chaturvedi “Acomprehensive survey on answer generation methods using NLP”,
ScienceDirect, Elsevier,2024.
2 The seminar content is based on the paper: "Generating Executable Test
Scenarios fromAutonomous Vehicle Disengagements Using Natural Language
Processing" by researchers at LundUniversity.
3 Scenario Generation based on Real World Accidents- IEEE Xplore.
4 Survey of Scenario Generation for Automated Vehicle Testing and Validation" -
ResearchGate,
5 Research on Specific Scenario Generation Methods for Autonomous Driving
Simulation Testing"- Mdpi
6 Jamin Rahman Jima,Md Apon RiazTalukderb,Partha Malakarb,Md Mohsin
Kabirc, Kamruddin Nurb,[Link], “Recent advancements and challenges of
NLP-based sentiment analysis: Astate-of-the-artreview”, ScienceDirect,
Elsevier,2024.
Dept of Comp Engg Short Title dd-October-2024 12 /
Dept of Comp Engg Short Title dd-October-2024 13 /
