From Coursera, Introduction to Self-Driving Cars by University of Toronto
https://www.coursera.org/specializations/self-driving-cars?action=enroll
The Requirements for Autonomy
Some terms and definitions:
Driving task:
- perceiving the environmrnt
- Planning how to reach from point A to B
- Controlling the vehicle
Operational Design Domain (ODD)
Taxonomy of Driving
How to classify driving system automation?
- Driver attention requirements
- Driver action requirements
- What exactly makes up a driving task?
- Lateral control: steering
- Longitudinal control: braking,accelerating
- Object and Event Detection and Response(OEDR):detection,reaction
- Planning: long term, short term
- Miscellaneous
Taxonomy:
- Level 0: no automation
- regular vehicles, no automation
- all be done by the driver
- Level 1: Driving Assistance
- either Longitudinal or Lateral control or both
- e.g.:
- Adaptive Cruise Control: can control speed, driver has to steer
- Lane Keeping Assistance: can help you stay in your lane,if you drift
- Level 2: Partial Driving Automation
- need both Longitudinal and Lateral Control
- e.g.:
- GM Super Cruise
- Nissan ProPilot Assist
- Level 3: Conditional Driving Automation
- Longitudinal + Lateral Control + OEOR
- Includes automated object and event detection and response (Key differece between level 2)
- e.g. Audi A8 Sedan
- Level 4: High Driving Automation
- Longitudinal + Lateral Control + OEOR + Fallback
- can handle emergencies autonomously, driver can entirely focus on other tasks. But may need driver to handle specific emergencies.
- e.g.: Waymo
- Level 5: High Driving Automation
- Longitudinal + Lateral Control + OEOR + Fallback + Unlirmited ODD
Requirement for perception
What is perception
- identification:
- to know what is the object in front
- understanding motion:
- know what it will do
Goals for perception
- Static:
- Road and lane markings(on-road)
- Curbs(off-road)
- Traffic lights(off-road)
- Road signs(off-road)
- Construction signs, obstructions, and more(on-road)
- dynamic objects (on-road): we need to predoct their motion
- Vehicles
- 4 wheelers(cars, truck…)
- 2 wheelers(motorbikes, bicycles…)
- Pedestrians
- Vehicles
- Ego requirements
- Position
- Velocity, acceleration
- Orientation, angular motion
Challenges to perception
- Robust detection and segmentation
- use Machine Learning, deep learning
- Snesor uncertainty: sensors may be suddenly broken
- Occlusion, reflection
- Illumination, lens flare
- Weather, precipitation
Driving decision and actions
Planning
Making decisions:
- Long term:
- how to navigate from New York to Los Angeles
- Short term:
- Can I change my lane to the lane right of me?
- Can I pass this intersection and join the left road?
- Immediate
- Can I stay on track on this curved road?
- Accelerate or brake, by how much?
Rule Based Planning
- What we just did was rule based planning
- Involved decision trees!
- In reactive rule based planning, we have rules that take into account the current state of ego and other objects and give decisions.
- Examples:
- If there is a pedestrian on the road, stop.
- If speed limit changes, adjust speed to match it.
Types of planning
- Predictive Planning:
- Make predictions about other vehicles and how they are moving.Then use these predictions to inform our decisions.
- Example:
- That car has been stopped for the last 10 seconds. It is going to be stopped for the next few seconds.
- relies on the accuracy of prediction
- Reactive planning