Human-Robot Interaction conference in Chicago, 2018

Posted on Sun 29 April 2018 in research

I attended the 2018 conference on Human-Robot Interaction. The primary reason was to present work done by VGIS students I had supervised during the autumn semester of 2017. Secondly, to gather inspiration for the new Grand Solutions - Augmented Cellular Meat Production (ACMP) project I was part of. Not that I would spend much more time on the HRI aspects but AAU had just acquired a new PhD that would work on this aspect of the ACMP project and I would like to be somewhat up-to-date on the topic.

General take-ways

  • Test on the actual end users - use user study experts for this
  • Precise knowledge of objects and actors locations is needed - how do we get that?
  • Collaborating with real industrial robots - No one at HRI was looking specifically at this
  • Possible collaborator in how robots are received in the workplace - Institut for Uddannelse og Pædagogik(http://edu.au.dk/)
  • VR is seen as mature, AR not yet but will be soon and is therefore also more interesting from research point of view
  • Nice explanations of mixed reality - "Making the hidden visible" or "exposing the hidden"
  • Nothing is completely autonomous - the degree of interaction just varies
  • Infrastructure is essential for efficient collaboration e.g. lane lines and signals
  • Adapt the systems to the expertise level of the user to improve efficiency and avoid frustration - measure expertise based on tracks (shows clear difference)

Keynotes

Steve Cousins' Keynote - Building a Service Robotic Business, Challenges from the Field

  • Use case is key, is it an aspirin or a vitamin
  • Robots help avoid monday models
  • Danger
  • Reduce impact of turnover
  • Other things than savings on workers hit the bottomline
  • Relay for transport between cells?

David Mindell's Keynote - Autonomy in human environments

  • What is autonomy? Never complete autonomy
  • Use of infrastructure to facilitate safe and efficient collaboration e.g. TLs People <-> robots <-> infrastructure <-> people
  • Detect expertise from precise tracks
  • Knowledge of precise location is essential

Specific Research

Titles and short summaries of presented works that would be worth looking further into.

  • Teaching users to remote control robot using scaffolding. dynamic curves controls the amount of scaffolding. Used to train a human operator, could it be used the other way?
  • "Its all in your head". Priming the expectations of the user. Large impact on the perception of the product.
  • Projecting robot intentions into the physical world. Highlighting the robots next move. Gamification.
  • Human-robot handover - Vary based on user experience level. Safety concerns solve by augmenting to show robot capabilities.
  • "A framework for robot generated mixed reality deixis" - New classes are possible in MR, annotations with e.g. arrows
  • "Transfer learning for robot navigation" - RNN learns to control robot through forrest. Sim-to-real.
  • "Training robots to acquire new skills" - RL PyBullet
  • Task planning in VR
  • Exposing the hidden
  • "Learning grasp from kinaesthetic demonstrations and mental simulations" - map between object and contact point, model hand/tool, evaluate grasp before execution
  • "Designing gripper by looking at what/how humans grasp" - record video of grasps, categorise, rank
  • "Teaching users to remote control robot using scaffolding" - dynamic curves controls the amount of scaffolding. Used to train a human operator, could it be used the other way?

its all in your head

Priming the expectations of the user Main poster session


pic1

"Virtual, Augmented, and Mixed Reality for Human-Robot Interaction" workshop poster session with VGIS students' poster

pic2