Dangers in Multiagent Rescue using DEFACTO Janusz Marecki

Dangers in Multiagent Rescue using DEFACTO Janusz Marecki Nathan Schurr, Milind Tambe, University of Southern California Paul Scerri Carnegie Mellon University

Dangers in Multiagent Rescue using DEFACTO Dangers in Multiagent Rescue l l Autonomous Multiagent Rescue – Problem: Which house to rescue first? – Human expertise & responsibility Human supervisor – 2 l Problem: Human overwhelmed with tasks Mixed decision making = DANGER ? ?

Dangers in Multiagent Rescue using DEFACTO Outline l l l 3 Motivation and Domain DEFACTO System Adjustable Autonomy Strategies Predicted results Experimental results & Dangers Summary

Dangers in Multiagent Rescue using DEFACTO Motivation l l 4 Large scale disasters Incident commander

Dangers in Multiagent Rescue using DEFACTO Domain timeline l l l 5 Currently: – Thorough testing of DEFACTO system Short term goal: – Los Angeles Fire Department Training Tool Long term goal: – Automated First Responders under human supervision

Dangers in Multiagent Rescue using DEFACTO Outline l l l 6 Motivation and Domain DEFACTO System Adjustable Autonomy Strategies Predicted results Experimental results & Dangers Summary

Dangers in Multiagent Rescue using DEFACTO System Architecture Demonstrating Effective Flexible Agent Coordination Through Omnipresence 7

Dangers in Multiagent Rescue using DEFACTO System Architecture Robocup Rescue Simulation Environment l 7 different simulators (fire, traffic, civilians etc. ) Demonstrating l Different maps (USC, Kobe) Effective Flexible Agent Coordination Through Omnipresence l 8

Dangers in Multiagent Rescue using DEFACTO System Architecture 9

Dangers in Multiagent Rescue using DEFACTO DAFACTO Movie 10

Dangers in Multiagent Rescue using DEFACTO System Architecture Simulator Fire. Brigade Machinetta Agent l 11 Fire. Brigade Machinetta Agent Machinetta: Multiagent platform, Abstracted Theories of Teamwork (Scerri et al AAMAS 03)

Dangers in Multiagent Rescue using DEFACTO Outline l l l 12 Motivation and Domain DEFACTO System Adjustable Autonomy Strategies Predicted results Experimental results & Dangers Summary

Dangers in Multiagent Rescue using DEFACTO Adjustable autonomy strategies l l 13 Agents dynamically adjust own level of autonomy – Agents act autonomously, but also. . . – Give up autonomy, transferring control to humans When to transfer decision-making control – Whenever human has superior expertise – Yet, do not overload human with tasks! – Previous: Individual agent-human interaction

Dangers in Multiagent Rescue using DEFACTO Team level Adjustable Autonomy l AT H AH l A TH l B l EQH l l 14 Team level A strategy Human strategy for all tasks Individual A strategy followed by the H strategy Team level A strategy followed by the H strategy The maximum number of agents the human is able to control The quality of human decisions

Dangers in Multiagent Rescue using DEFACTO Outline l l l 15 Motivation and Domain DEFACTO System Adjustable Autonomy Strategies Predicted results Experimental results & Dangers Summary

Dangers in Multiagent Rescue using DEFACTO Calculating predictions l l 16 Strategy value equations Domain specific

Dangers in Multiagent Rescue using DEFACTO Predicted results Low B, Low EQh l 17 Low B, High EQh Although higher expected quality of human decisions yields better results, low limit of human controllable agents hampers the overall score

Dangers in Multiagent Rescue using DEFACTO Predicted results - ctnd High B, Low EQh l 18 High B, High EQh High limit of human controllable agents makes the human involving strategies effective also for larger teams, beating the fully autonomous A strategy

Dangers in Multiagent Rescue using DEFACTO Outline l l l 19 Motivation and Domain DEFACTO System Adjustable Autonomy Strategies Predicted results Experimental results & Dangers Summary

Dangers in Multiagent Rescue using DEFACTO Experimental setup l l l 20 3 Subjects Allocation Viewer Same Map for each scenario – Building size and location – Initial position of fires 4, 6, and 10 agents A, H, ATH Strategies Averaged over 3 runs

Dangers in Multiagent Rescue using DEFACTO Experimental results 21

Dangers in Multiagent Rescue using DEFACTO Conclusions from results l l 22 l No strategy dominates through all the experiments in all cases As the number of agents increase, for strategy A the slope of improvement is greater than the slope of improvement for H. This correlates with our prediction that humans are not as good at exploiting additional agents resources, whereas agents are able to better exploit increasing numbers of available teammates If the difference for 4 agents between strategy A and H for a particular commander is small enough, as is the case with subjects A and C, then as we grow to larger numbers of agents, A will dominate AH, ATH and H ATH was constructed to help out at large # of agents in the team. However, what we see instead is that ATH does better at smaller # of agents over H, in a very surprising result. At higher # of agents, ATH does worse for subject A than A. Dip at 6 agents?

Dangers in Multiagent Rescue using DEFACTO Discrepancy for 6 agents? l l 23 At 6 agents case, mixed strategies involving humans and agents (AH and ATH) performed worse than for 4 agents case At 6 agents case, H strategy improved over the 4 agents case At 6 agents case, AT strategy improved over the 4 agents case Hypothesis: Human-Agent conflicts in resource allocation caused the problem

Dangers in Multiagent Rescue using DEFACTO Task allocation overload danger 24

Dangers in Multiagent Rescue using DEFACTO Summary l l l 25 Rigid transfer of control strategies are outperformed by flexible dominant strategy selection Having human in the loop does not necessary lead to increased performance Having humans and agents doing resource allocation simultaneously is susceptible to excessive reallocations which decreases overall performance

Dangers in Multiagent Rescue using DEFACTO Future application l 26 Automated First Responders using DEFACTO

Dangers in Multiagent Rescue using DEFACTO Thank you! l l l Email: marecki@usc. edu Teamcore web site: http: //teamcore. usc. edu Thanks – – 27 CREATE Center Fred Pighin, Pratik Patil, Nikhil Kasinadhuni and J. P. Lewis