Intelligent Overspeed Control in Autonomous Vehicles with DQN Deep Reinforcement Learning.
Date:
In this talk, we explore the implementation of our research titled “Intelligent Overspeed Control in Autonomous Vehicles with DQN Deep Reinforcement Learning.”
Introduction
- Road accidents cause about 3700 deaths per day (WHO)
- Reckless driving and over-speeding are major causes
- Different zones require different speed limits
- Residential
- School zones
- Highways
- Need intelligent speed control system
Research Background
- Traditional speed control lacks adaptability
- Reinforcement Learning learns from environment interaction
- Deep Reinforcement Learning improves performance
- Research gap:
- Limited work on zone-based intelligent speed control
Research Objective
- Develop a DQN-based intelligent speed control system
- Detect over-speeding in different zones
- Apply intelligent deceleration:
- High
- Medium
- Low
- Optimize vehicle speed according to zone limits
Methodology
- 4×4 grid environment simulation
- Zone-based speed limits (20–80 km/h)
- Inputs:
- Current speed
- Zone speed limit
- Speed difference
- Action space:
- High deceleration
- Medium deceleration
- Low deceleration
- Reward function:
- Encourage safe speed
System Architecture
- Environment state input
- LSTM-based DQN model
- Action selection mechanism
- Speed adjustment output
Key Findings
- Model adapts to different speed zones
- Intelligent deceleration improves safety
- Effective decision-making under varying conditions
Discussion & Implications
- Reduces over-speeding in multiple zones
- Adapts to uncertain driving environments
- Supports autonomous and semi-autonomous driving safety
Conclusion
- DQN-based speed control is effective
- LSTM-DQN selects appropriate deceleration
- Dynamic speed control improves road safety
Future Work
- Use real-world driving datasets
- Multi-agent learning for traffic scenarios
- Add pedestrian and collision detection
- Real-time deployment in autonomous vehicles
Thank You
Feel free to contribute, raise issues, or suggest improvements!
After Technical Session
