Carnegie Mellon University THE ROBOTICS INSTITUTE

Thesis ProposalJonathan Michael Butzke

Wednesday, December 2, 2015 GHC 810212:00 p.m.

Maxim Likhachev Chair

Sebastian Scherer

Nathan Michael

Dan Lee University of Pennsylvania

Thesis Committee

Planning for a Small Team of Heterogeneous Robots:from Collaborative Exploration to Collaborative Localization

Abstract Robots  have  become  increasingly  adept  at  performing  a  wide  variety  of  tasks  in  the  world.  However,  many  of  these  tasks  can  benefit  tremendously  from  having  more  than  a  single  robot  simultaneously  working  on  the  problem.  Mul>ple  robots  can  aid  in  a  search  and  rescue  mission  each  scou>ng  a  subsec>on  of  the  en>re  area  in  order  to  cover  it  quicker  than  a  single  robot  could.  Alterna>vely,  robots  with  different  abili>es  can  collaborate  in  order  to  achieve  goals  that  individually  would  be  more  difficult,  if  not  impossible,  to  achieve.  For  example,  in  an  explora>on  scenario,  a  ground  robot  can  carry  a  large  baEery  capacity  providing  it  with  a  higher  endurance,  and  thus   could   spend   longer   searching   than   an   aerial   vehicle.   Conversely,   the   ground   vehicle   by   its   very   nature   is  more   limited   in   the  terrain  that  it  can  traverse  than  the  aerial  vehicle.  In  order  to  adequately  explore  a  large,  obstacle  strewn  environment,  the  two  robots  will  have  to  intelligently  determine  when  is  the  best  >me  and  where  is  the  best  posi>on  to  employ  each  of  them.  In  these  cases,  mul>-­‐robot   collabora>on   can   provide   benefits   in   terms   of   shortening   search   >mes,   providing   a   larger   mix   of   sensing,   compu>ng,   and  manipula>on   capabili>es,   or   providing   redundancy   to   the   system   for   communica>ons   or   mission   accomplishment.   One   principle  drawback  of  mul>-­‐robot  systems  is  how  to  efficiently  and  effec>vely  generate  plans  that  use  each  of  the  team  members  to  their  fullest  extent,  par>cularly  with  a  heterogeneous  mix  of  capabili>es.  

Towards   this   goal,   I   have   developed   a   series   of   planning   algorithms   that   incorporate   this   collabora>on   into   the   planning   process.  Star>ng   with   systems   that   use   collabora>on   in   an   explora>on   task   I   show   teams   of   heterogeneous   ground   robots   planning   to  efficiently  explore  an   ini>ally  unknown  space.  These  robots  share  map   informa>on  and   in  a  centralized  fashion  determine  the  best  goal  loca>on  for  each  taking  into  account  the  informa>on  gained  by  other  robots  as  they  move.  This  work  is  followed  up  with  a  similar  explora>on   scheme  but   this  >me  expanded   to   an   air-­‐ground   robot   team  opera>ng   in   a   full   3-­‐dimensional   environment.   The  extra  dimension  adds  the  requirement  for  the  robots  to  reason  about  what  por>ons  of  the  environment  they  can  sense  during  the  planning  process.  With  an  air-­‐ground  team,  there  are  por>ons  of  the  environment  that  can  only  be  sensed  by  one  of  the  two  robots  and  that  informa>on  informs  the  algorithm  during  the  planning  process.  Finally,  I  extend  the  air-­‐ground  robot  team  to  moving  beyond  merely  collabora>vely   construc>ng   the   map   to   actually   using   the   other   robots   to   provide   pose   informa>on   for   the   sensor   and  computa>onally   limited   team   members.   By   explicitly   reasoning   about   when   and   where   the   robots   must   collaborate   during   the  planning  process,  this  approach  can  generate  trajectories  that  are  not  possible  if  planning  occurred  on  an  individual  robot  basis.