Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Active Modular Environment for Robot Navigation

Active Modular Environment for Robot Navigation

More Decks by Keisuke Okumura | 奥村圭祐

Other Decks in Research

Transcript

  1. Active Modular Environment for Robot Navigation Shota Kameyama*, Keisuke Okumura,

    Yasumasa Tamura & Xavier Defago May. 30 – Jun. 5, 2021 Tokyo Institute of Technology, Japan ౦ژ޻ۀେֶ 5PLZP*OTUJUVUFPG5FDIOPMPHZ virtual conf. (Xi’an) ICRA-21 *graduated, now at Sony Corp.
  2. /15 2 navigation internal representation ≠ external environment e.g., sensor

    noise, dynamic env. Representation & Planning time-loss or collisions planning representation robot unexpected accidents
  3. /15 3 Our Concept navigation planning representation robot offload robots

    have neither representation nor planning function active environment computing unit “cells”
  4. /15 4 Benefits of Offloading Representation & Planning robots =>

    environment 1. functional separation no burden of collecting info and planning trajectories 3. multi-robot coordination collision avoidance is achieved by environment 2. response to dynamic environment save time for replanning
  5. /15 5 cells plan info ours planning with explicit representation

    sensing classical navigation no sensing no representation no planning [Visvanathan+ SENSORS-15, Sinha+ ICCIC-17] planner info plan representation remote control no representation no use of global info no centralized unit sensor/tag (passive env.) info planning without explicit representation stigmergy [Kim & Chong TAS-08, Khaliq&Saffiotti ICRA-15, etc] no planning no collective behaviors What this is NOT
  6. /15 10 How Cells Coordinate path planner topology manager physical

    layer run multiple distributed algorithms concurrently and hierarchically self-stabilizing distributed routing c.f., [Tel 00, Johnson & Mitra TCS-15] collision avoidance, reservation protocol +connection detector, message controller, etc
  7. /15 13 Navigation in Dynamic Env. 30 repetitions round trip

    several times stochastically becoming obstacles AFADA self-nav recovery rate 0 10 20 30 40 1% low 5% high simple-loop 0 20 40 60 80 100 120 1% 5% two-bridge 0 20 40 60 80 100 1% 5% two-loops steps little useless moves * ** **
  8. /15 14 As Platform of Multi-Robot Coordination multi-robot navigation in

    decentralized style with offline centrealized plan [Okumura+ preprint-21] automated parking 6x4, 8 robots 8x8, 10 robots
  9. /15 15 Concluding Remarks AFADA; active modular environment, working as

    infrastructure proof-of-concept offloading representation & planning from robots to env. concept smooth integration representation, planning, and execution motivation future directions c.f., [Thomas+ Intell. Syst.-15, Li+ AAAI-19, Atzmon+ SoCS-19] large robots cell-driven multi-robot path planning c.f., [Okumura+ AAAI-21] video&authors: https://dfg-lab.github.io/afada/