Guidelines for Use
Use for benign purposes only. While controlling Neatos might be relatively harmless, applying the movement code to bigger and heavier robots has the potential to go wrong. If the robot misinterprets a direction, it could fall on or hit a human. To ensure that the robot will not accidentally cause harm, prevent people from getting close to the robot when it is running.
Intended Use Cases
Making robots dance/non-harmful actions
Robots must be in a controlled space
Must inform people in the vicinity
Prohibited Use Cases
Weapons platform of any sort
Causing harm to any living beings
Surveilling people without their knowledge or consent
Interactions with CVPR's Potential Negative Societal impacts
Directly facilitate injury to living beings. For example: could it be integrated into weapons or weapons systems?
Yes. The robot could be controlled to attack someone. A weapon could be additionally be integrated into the system with a trigger. However it should be noted, that a significant amount of work to code the robot would need to be done for that to happen.
Raise safety, privacy, or security concerns.
Potentially. The webcam on the computer is used to collect joint information on the person controlling it. However it is not stored and all joint information that is streamed to the robot is encoded as movement commands.
Raise human rights concerns.
Potentially. The system uses an open source detection module called YOLO to do joint detection. We have not checked that YOLO has been trained on unbiased data or that it will work equally on all types of people.
Have a detrimental effect on people’s livelihood or economic security.
Unlikely. Because this robot needs a human to operate it.
Develop or extend harmful forms of surveillance.
Yes. While using data from joint positions to dance is relatively harmless, there could be other applications of using joint tracking. For example, it could be used to track what actions a person is taking and that information could be used in a harmful way.
Severely damage the environment.
Unlikely. This robot currently does not interact with the outdoors.
7. Deceive people in ways that cause harm.
Potentially. If the robot is advertised as a dancing robot, but the person controlling it wants to cause harm it could deceive people.
Human-derived Data Ethic Interactions
Contain any personally identifiable information or sensitive personally identifiable information.
Unlikely. People are not kept track of or labeled in this system.
Contain information that could be deduced about individuals that they have not consented to share.
Unlikely, again since individuals are not labeled or tracked.
Encode, contain, or potentially exacerbate bias against people of a certain gender, race, sexuality, or who have other protected characteristics.
Unknown, because we are using a model that has been trained on the COCO dataset. It seems, based on research others have done, that it has a severe gender bias, but we're unsure what other biases it might have.
Contain human subject experimentation and whether it has been reviewed and approved by a relevant oversight board.
Does not contain human subject experimentation.
Have been discredited by the creators.
No, it has not been.