the purpose of hart (hedonism, autonomy, responsibility, trust) niranjan suri maria gini, seyed...

13
The Purpose of HART (Hedonism, Autonomy, Responsibility, Trust) Maria Gini, Seyed Waqar Jaffry, Niranjan Suri Niranjan Suri , Janneke van der Zwaan, Arnoud Visser

Upload: cecil-simon

Post on 16-Dec-2015

220 views

Category:

Documents


1 download

TRANSCRIPT

The Purpose of HART(Hedonism, Autonomy, Responsibility, Trust)

Maria Gini, Seyed Waqar Jaffry, Niranjan SuriNiranjan Suri,Janneke van der Zwaan, Arnoud Visser

Long list issues (purpose related) Killer applications (different team sizes) What is teamwork? Role of communication in teamwork (reduction inside good team) Role of training inside team How to communicate H->R, R->H, R->R. Role of none-robot agents inside teams Multiple representation (both HAR) in same role How can RA be aware of the human and his feelings? Metrics; teams, performance, meta-information about the experiments,

quantitative and qualitative methodologies Emotions for the robots to support the group interactions Transparency / deceiving Lifetime of a team (team formation / role taking / specialization /

breaking up) Training is partly done by the environment / partly from the team-

members Do robots still need a identity for scale > 100 Explore / control / autonomy -> gradual responsibility

Short list issues (purpose related) Metrics; teams, performance, meta-information about the

experiments, quantitative and qualitative methodologies 1, 2

How can RA be aware of the human and his feelings? 2, 2 Lifetime of a team (team formation / role taking /

specialization / breaking up) 1, 3

Killer applications (different team sizes) 2, 3 How to communicate H->R, R->H, R->R. 1 Do robots still need a identity for scale > 100 1

Multiple representation (both HAR) in same role 3 Emotions for the robots to support the group interactions 3

Short list issues (purpose related) Metrics; teams, performance, meta-information about the

experiments, quantitative and qualitative methodologies 1, 2

How can RA be aware of the human and his feelings? 2, 2 Lifetime of a team (team formation / role taking /

specialization / breaking up) 1, 3

Killer applications (different team sizes) -> Milestones How to communicate H->R, R->H, R->R. 1 Do robots still need a identity for scale > 100 1

Multiple representation (both HAR) in same role 3 Emotions for the robots to support the group interactions 3

Milestone 1&2 (killer applications) Example of team of 10 (2 drivers / 8 robots)

Convoy of trucks with two humans:easier on the highway than behind the warfront

Traffic lights should be intelligent before the convoy isallowed on highway

The army application should be operational in 2016

Ethics / laws will be the main issue to get this scenario operational before 2020 at the highways

Milestone 3 (killer applications) Example of team of 2 (1 nurse / 1 robot)

Lifting assistance for a nurse

Alternatives: Making the bed more intelligent (too restrictive) Get the lifting capability in the environment (a lot of unfriendly infrastructure) Exoskelet for the nurse (unfriendly for the other tasks of the nurse)Lifting robot is most flexible and friendly solution

The challenge of lifting is still at the mechanical department. Should we already work on the teamwork?

The robot should automatically follow the nurse and keep out of the nurse’s way

The robot should know when to follow and when to stay (judging the intentions)

Milestone at 2020

Milestone 4 (killer applications) Example of team of 5 (2 operators / 1 human explorer / 1 ground robot / 1

air robot)

Urban Search and Rescue

Milestone January 2011

Trust Hard to gain, easy to lose

How do robots gain and maintain human trust Do robots need to trust humans?

Predictability of system is key to trust Prescriptive notion – if the system is performing as specified by the

objective, then it can be trusted Another indicator – “normal” communication between team

members Effective conveyance of performance and limitations to humans How does the human recognize and convey to the robot

detrimental environmental conditions Analogy to Coaching – need to understand the limits, why something is

going wrong, and convey options / direction to the robots Dynamics of Human Trust – key role in team building Teaming is something that is perceived by humans, not the robots Robots must adapt their behavior based on their perception of the

human trust model Robots must be able to interpret indirect human expression /

communication

Autonomy and Responsibility “A Robot without a battery is a very autonomous robot” – Visser “A Robot that can say no is a very autonomous robot” - Gini Meta question – what qualifies as a robot? Is Autonomy Required?

Bulldozer?, Robonaut? Jackhammer? Car? Behavior-based Robotics – two goals:

Invariants that should not be violated (do not run into walls, kill anyone, including self, etc.)

Attainable goals – related to task or objective How does teaching (e.g., by demonstration) / learning fit into this?

How does learning for a robot differ from learning for humans? Awareness of the human team member(s) is paramount To what extent, if at all, do we need to anthropomorphize robots? Not

just from a physical perspective, but from the notions of trust, behavior, etc.

Responsibility – four types? Responsibility for the given / assigned / delegated tasks Responsibility to communicate with team members for shared SA Responsibility to the other team members / task (e.g., help other team

members) Responsibility for the greater good (e.g., not polluting the environment, etc.)

Applications Search and Rescue

UAV – Mountain Search (Brown University) RoboCup City-Level Search and Rescue for Disaster Recovery

Persistent Surveillance Fixed and mobile assets

Cooperative Medicine Nurse’s aid Operating room Rehabilitation

Assistants for Elderly Citizens Warfighting Useful to categorize applications into good targets for teams

of size 2, 5, and 10 members Nurse’s aid, Elderly Citizen Assistants (2) Mountain Search and Rescue, Persistent Surveillance (5) Warfighting (10)

Metrics What are appropriate metrics to measure success of

teamwork? We don’t want a Turing Test, but some test One Approach is to show that Human + Robot can do more than

Human + Human or Robot + Robot Scalability with respect to numbers

What are the best domains to show HART? Meta metric – how “much” teamwork is there in a solution? Metric – how well is the solution working? Is there a notion of a local (i.e., individual) objective or metric

versus a global (i.e., systemwide) objective or metric? How does robustness play into this? Should robustness be an

independent metric? “Types” of robustness – flexibility to deal with novel/different

situations (opposite of brittleness) Role substitution – humanoid robots substituting for humans Fit – how well do robots fit in an environment constructed by and

for humans

Questions What is essential for teamwork?

Human-level communication (gestures, facial expressions)?