elective in artificial intelligenceiocchi/didattica/hri/slides/2017/hri_2a... · the hatp...
TRANSCRIPT
Human-Robot Interaction
Elective in Artificial Intelligence
Lecture 2 – Interaction Management
Luca Iocchi
DIAG, Sapienza University of Rome, Italy
M. Cirillo, L. Karlsson, A. Saffiotti. Human-aware task planning: An application to mobile robots. In ACM Transactions on Intelligent Systems and Technology (TIST), 1 (2), 2010.
de Silva, L. and Lallement, R. and Alami, R. The HATP Hierarchical Planner: Formalisation and an Initial Study of its Usability and Practicality. In Proc. of the IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS), 2015.
Valerio Sanelli, Michael Cashmore, Daniele Magazzeni, and Luca Iocchi.Short-Term Human Robot Interaction through Conditional Planning and Execution.In Proc. of International Conference on Automated Planning and Scheduling (ICAPS) 2017 (to appear).
Eugenio Sebastiani, Raphaël Lallement, Rachid Alami, and Luca Iocchi. Dealing with On-line Human-Robot Negotiations in Hierarchical Agent-based Task Planner.In Proc. of International Conference on Automated Planning and Scheduling (ICAPS) 2017 (to appear).
L. Iocchi - Human-Robot Interaction 2
Readings
Interaction design/management
• Efficiency and effectiveness of “interaction work”
Human-Robot Interaction design/management
• Efficiency and effectiveness of HRI
L. Iocchi - Human-Robot Interaction 3
HR Interaction Design
Interaction = execution of actions by humans and robots according to the current situation
Interaction Design = Action planningInteraction Management = Plan execution
Formalisms & Tools
• Low-level (encoding interaction in the program/behavior)
• High-level (explicit representation of the interaction)
L. Iocchi - Human-Robot Interaction 4
HRI Design as Planning
High-level frameworks for HRI Design and Management
• HA-PTLplan – Human-aware Task Planner [Cirillo et a., 2010]
• HATP – Hierarchical Agent-based Task Planner [Lallement et al. 2014, de Silva et al. 2015]
• HATP + Sensing [Sebastiani et al. ICAPS 2017 (to appear)]
• ROSPlan conditional planning[Sanelli et al. ICAPS 2017 (to appear)
• MDP + Sensing [Iocchi et al. ICAPS 2016]
L. Iocchi - Human-Robot Interaction 5
HR Interaction Design
[Cirillo et al., 2010]• Robots take into account human actions both at planning time and at
execution time. Does not plan for human actions.• Respecting interaction constraints.• Multiple hypotheses about the world (Partial Observability/POMDP)
L. Iocchi - Human-Robot Interaction 6
Human-aware Task Planning
Human actions predicted (env. sensors / HMM / plan recognition)
Monitor / Failure detection / Replanning
L. Iocchi - Human-Robot Interaction 7
Human-aware Task Planning
[Lallement et al. 2014, de Silvae et al. 2015]• Based on Hierarchical Task Networks (HTN)
• Distinguish different agents
• Generates different streams of actions (linear plans) for each agent
• Generate plans also for humans.
L. Iocchi - Human-Robot Interaction 8
Hierarchical Agent-based Task Planner
Limitations
• Roles of humans and robots are decided at planning time
• Humans must be aware of the plan and execute it according to the given constraints
• No flexibility in on-line modification and adaptation
L. Iocchi - Human-Robot Interaction 9
Hierarchical Agent-based Task Planner
kcl-planning.github.io/ROSPlan
Integrate planners in ROS
Modular and extendible
L. Iocchi - Human-Robot Interaction 10
ROSPlan
Plan representation• Sequence of actions • Temporal model• Petri Nets
Execution mechanism• Continuous state estimation • Re-planning
L. Iocchi - Human-Robot Interaction 11
ROSPlan
Classical/HTN Planning (e.g., STRIPS, HATP)• Complete knowledge about the state
• Output is a linear plan (sequence of actions)
• No representation of uncertainty
MDP/POMDP Planning• No explicit representation of sensing actions
• Assumes perfect or very reliable perception
L. Iocchi - Human-Robot Interaction 12
Discussion
• Execution mechanism• State estimation error-prone (perception involving humans
more difficult)
• Monitor + replanning after failures not adequate for HRI tasks
• Non-expert users highly unpredictable
L. Iocchi - Human-Robot Interaction 13
Discussion
Service robot has to make sure user needs are satisfied. User needs are not known in advance.
L. Iocchi - Human-Robot Interaction 14
Example
Classical planning
• Guess a user need
• Plan with this guess
• Execute the plan
• If guess is wrong, adjust conditions and replan
When guess is wrong, behavior is not socially acceptable.
• The robot does not move, but the user needs something.
• The robot prepares food or drink that is not requested by the user.
L. Iocchi - Human-Robot Interaction 15
Example
Plan with explicit sensing action
• Go to person
• Ask if s/he needs something // Sensing action
• if (need_food AND need_drink)• Go to the kitchen
• Prepare food and drink
• Serve food and drink to person
• if (need_food)• …
• else• Do nothing
L. Iocchi - Human-Robot Interaction 16
Example