A Safety Architecture for Autonomous Agricultural Vehicles David - - PowerPoint PPT Presentation

A Safety Architecture for Autonomous Agricultural Vehicles

David Kohanbash Carnegie Mellon University Marcel Bergerman Carnegie Mellon University Karen M. Lewis Washington State University Stewart J. Moorehead John Deere & Co. dkohanba@cmu.edu IEEE AgRA January 27, 2014

Why Robots

http://www.census.gov/population/international/data/idb/images/worldpop.png

Why Robots

Source: Rodney Brooks talk http://www.youtube.com/watch?v=nA-J0510Pxs

Why Robots

Source: Rodney Brooks talk http://www.youtube.com/watch?v=nA-J0510Pxs

• Robotics can significantly:

– Reduce labor demand – Reduce labor (and other) costs – Increase efficiency

Current Robots

• Industrial robots have been a crucial part of

society for the last 60 years

– They operate safely (as long as people are not around), reliably, and do not adept well to change – The user interface does not require intimate knowledge to use the machine – Safety regulations and standards have led safe machine creation

Agricultural Robots

• Large
• Fast
• Changing Environment
• People
• Easy to reconfigure
• Mowing
• Spraying
• Disease Detection
• Pest Detection
• Yield Estimation
• Pruning
• Harvesting
• Transporting

Sensing, Computing, & Actuation

• Safety Assessments (pre-mortems)
• Fail Safely
• Appropriate emergency-stops
• Autonomy system can not interfere with safety

systems

• Detect component failures
• Avoid damaging people, infrastructure, and crops
• Third party verification of safety systems

Human-Machine Interface

• Easy to use
• Warning lights/sound
• Intuitive
• Training
• Process control (safety vests, signage, etc..)

Standards & Regulations

Legal frameworks for testing (and deploying) cars that drive by themselves in Nevada, Florida, California, and the UK..

Google self driving car

http://www.geek.com/articles/news/google-wants-self-driving-cars-in-nevada-20110512/google-self-driving-car/ http://www.insideline.com/toyota/prius/blind-man-tests-googles-self-driving-car-on-california-roads.html

• Operator must approve

plan and modifications to plan

• Operator must approve initial plan
• Operator should be able to remotely view and

modify vehicle plans

• Must be tested monthly
• - 10, 000 testing hours required before vehicle

model can be used unmanned

• - System in place to avoid remote tampering

including wifi and GPS

• - Vehicle must be able to handle being GPS

denied

• Vehicles should be able to

communicate paths with each

• ther when working in close

proximity

• - Vehicles should notify
• perator when entering a new

zone

• Vehicle should alert driver

before maneuvering away from an obstacle

• - User should be able to

take control by grabbing steering wheel or pedals

• Must be able to detect people in various

positions (standing, crouched, laying, operating) with over 50% person occlusion

• Stop if obstacle is potentially a person and do

not try to route around

• non-autonomous vehicles get right of way
• - Must be tested daily
• Vehicles should not maneuver

around other vehicles without

• perator approval
• Deadman switch
• -Vehicle should stop if

perception system degraded

• Systems in place to avoid

tampering

• 360degree image view for remote operators
• Vehicle should be able to completely stop in x

seconds of obstacle detection or user command

• - Real time user communication with less than X

seconds must be maintained

• - Vehicle should have warning lights and sirens.

Must begin 1 minute before any motion

• - Must be tested daily
• - Develop system for establishing vehicle keep-
• ut zones
• Operator should be able to

view any remote vehicle

• - Maximum multi-hop

communication jump can not exceed X vehicles

• - Provide a method to stop all

vehicles.

Future Needs

• Focus on not just building a system but

building a safe system

• Attention to user interface
• More standards to guide autonomous (or

semi-autonomous) vehicles

Thank You Lets start the discussion…

This work is partially supported by the US Dept. of Agriculture National Institute for Food and Agriculture under award nos. 2008-51180-04876 and 2008-51180- 04890.The Autonomous Orchard Tractors were funded internally by John Deere.