It could be argued that at present robot technology has reached a point where it could be used in peoples homes. However, there are still no methods for certifying a robot as safe to work with people.
The main questions therefore are: * Can we prove that a robot will behave safely? * What approaches are there to certify a robot as safe?
This is an on-going topic of discussion in the robotics research community. To date, there has been a number of people working on this problem. For example, one of Willow Garage's robots (PR2) are built with arms based on a paper by Wyrobek et al., called "Towards a Personal Robotics Development Platform: Rationale and Design of an Intrinsically Safe Personal Robot". This and other work tackles the problem of 'how to build a robot that is physical safe for a human user to interact with it?', but whether a robot's 'decision making' will result in a safe outcome for a human user falls into the domain of roboethics. For further discussion on roboethics, please visit the Wikipedia and/or Roboethics Info DataBase (www.roboethicsDB.com).
I would be very much interested to find out more about Razor Robotics' approach to building safe robots. :)
The safety of Human-Robot Interaction is the main research aim of Razor Robotics. Coincidently we have applied for a PR2 Robot from Willow garage as part of their Beta testing program.
We are currently investigating a number of different areas of robot safety. The first area is primarily concerned with identifying visual cues from the Human user and trying to take this data and determine the intentions and eventual actions of the user.
The other area of research is looking into the problem of testing complex robot systems to ensure that there are no routines which may cause the robot to behave un-safely and minimising the effect of any such un-safe operations. We will be releasing a paper on this subject in the middle of May and I will make sure I post a link to this discussion and also on our safety pages.
Thank you so much for the quick reply. I can't believe how relevant your statement "The first area is primarily concerned with identifying visual cues from the Human user and trying to take this data and determine the intentions and eventual actions of the user." is to my research. I am looking at the reverse side of safety where the use of nonverbal gestures by a robot can be used to convey the robot's intentions to a human collaborator for more safe HRI.
I would definitely be interested to read the paper. Is it, by any chance, going to be presented at ICRA 2010 conference at all? It is being held in May, and I will be presenting my work there.
For details of my thesis work, please visit the 'Current Research' section of http://profile.amoon.ca/
The research is taking place at the Bristol Robotics Laboratory. My research is running alongside the Cooperative Human Robot Interaction Systems (CHRIS) project; details of which can be found here: www.chrisfp7.eu.
The paper will be presented at the DRHE 2010 Workshop. I will look out for your work when the ICRA 2010 conference journal is released :)
I guess it would be the logical assumption to say that a modern day robot would be safe in the house. It's not like we're at a point where it's simple AI would suddenly go rogue and attack, say, the old human being it's supposed to be taking care of?
Then again, wouldn't it also depend on what the robot was designed for? I mean, you could design a robot meant for washing the dishes in someones house and people probably wouldn't get too frightened. But if you had a robot designed for baby sitting, I'm sure people would be more then frightened to have a robot washing there child in a bathtub.
Even if the robot was designed to be water proof, people are going to be afraid of that fatal accident happening when something malfunctions and the robot accidentally electrocutes the kid in the water.
When I talk about robots working with people, I'm generally referring to robots with appendages fitted with manipulation devices. You are definitely correct that we are not at the stage where an Artificial Intelligence controller would go 'rogue'. However, designers of all safety-critical systems (like robots) are still faced (even after decades of software development) with the problem of system testing. As a system, like a personal robot is hugely complex, it is difficult (and currently impossible) to test all system eventualities. Therefore, if you want your robot to do flexible tasks, such as help you prepare a meal or fold laundry, there is no guarantee that the robot will always behave safely in all situations. This is due to the fact that currently robot systems are unable to identify the majority of safety risks and equally are unable to decide on the best actions to avoid or limit these risks.
To summarise, the main safety problems associated with personal robots are:
* Difficulty to test and verify robot controller (often referred to as 'The computational problem of system testing' * Full sensor coverage of the robot (replicating the array of sensors that humans have) * Understanding and usage of sensor readings (Perceiving the world) * Identifying potential safety risks * Correctly avoiding or limiting safety risks
Contact us 
About us 
Help 
