1 of 13

FINAL REVIEW MEETING

Luka Peternel

Cognitive Robotics, 3mE

Delft University of Technology

The Netherlands

​

​

ESA Rhizome project

Revision date: 22.5.2022

L.Peternel@tudelft.nl

2 of 13

OVERVIEW OF WORK

The Human-Robot Interaction (HRI) part involved three main aspects:

  1. Designing a controller for physical HRI during a pick-and-place task (contributor: Hugo Loopik).
    • This task is crucial for building a habitat and includes several key sub-tasks (e.g., carrying).
  2. Designing a trajectory learning and optimization based on human preferences (contributor: Armin Avaei)
    • Since the construction site is unstructured and unpredictable it is important for the robot to be able to learn from the collaborating human.
  3. Conducting a feasibility study for attaining robot mobility.
    • This conceptual study examined different mobile platforms and their power supply.

​

​

L.Peternel@tudelft.nl

3 of 13

CONTROL OF PHYSICAL HUMAN-ROBOT INTERACTION

L.Peternel@tudelft.nl

4 of 13

CONTROL METHOD FOR HRI

  • The developed control method for HRI consists of five modes that cover the requirements of the four identified sub-tasks:
    • Pick-up
    • Carrying
    • Alignment
    • Stand-by
  • To enable the human to switch between these modes we developed a voice interface based on the recognition of language commands.
  • The five designed modes of operation are:
    • Locked
    • Free
    • Main
    • Orientation
    • Lift & lower

L.Peternel@tudelft.nl

5 of 13

EXPERIMENT: LARGE OBJECT CARRYING

L.Peternel@tudelft.nl

6 of 13

TRAJECTORY LEARNING AND OPTIMIZATION�BASED ON HUMAN PREFERENCES

L.Peternel@tudelft.nl

7 of 13

LEARNING METHOD

  • We developed a method based on machine learning and optimization that can incorporate human preferences.
  • The method uses human demonstrations to infer the preferences, which are extracted from the measured data using the inverse reinforcement learning (IRL) approach.
  • We identified four main preferences that are fundamental to the construction task:
    • Carrying velocity
    • Height from the ground during the carrying
    • Minimum distance to obstacles during the carrying
    • Side on which the obstacle is passed.

L.Peternel@tudelft.nl

8 of 13

EXPERIMENT: LEARNING HUMAN PREFERENCES

L.Peternel@tudelft.nl

9 of 13

TOWARDS MOBILITY OF COLLABORATIVE ROBOT

L.Peternel@tudelft.nl

10 of 13

FEASIBILITY STUDY FOR MOBILE PLATFORM

​

  • So far, we have done experiments on a fixed-base collaborative robotic arm.

​

  • The constriction site has a large workspace and collaborative robots require a larger range.

​

  • It is vital that collaborative robots have sufficient mobility.

​

  • What are the best options for mobile platforms to carry the collaborative robotic arm?

​

L.Peternel@tudelft.nl

11 of 13

WHAT KIND OF MOBILE PLATFORM

  • Collaborative robotic arm (study case: KUKA LBR iiwa 14)
    • Arm mass: 29.5 kg (payload of 14 kg)
    • Robotic gripper: 5 kg (reduces arm payload to 9 kg)
    • Controller cabinet mass + accessories: 25.5 kg

​

  • Legged mobile platforms (e.g., humanoid robots, quadrupeds) are complex, less energy efficient and have limited payload.
  • The existing commercially available wheeled mobile platforms provide a good starting point; however, they are designed to operate in factory environments (e.g., no rough terrain, no slopes, etc.).

​

  • We would prefer a custom-designed mobile platform to be robust enough to operate in a rough environment:
    • Wheeled with additional “passive” metal legs for fixation
    • Number of wheels, design and configuration to be determined
    • Desired payload (in addition to carrying itself): 60 kg

​

KUKA mobile platforms (Kim et al, 2019)

Robotnik mobile platforms (Kim et al, 2020)

L.Peternel@tudelft.nl

12 of 13

BATTERY VS CABLE

  • Battery-powered robotic platform advantages:
    • No obstructions by the cable
    • No workspace limitations
  • Battery-powered robotic platform disadvantages:
    • Shorter work time and requires recharging (could use spare batteries→ extra mass for the mission and replacing time)
    • Extra load to be carried by the rover

​

  • Cable-powered robotic platform is advantages:
    • Unlimited work time
    • Less load to be carried by the rover
  • Cable-powered robotic platform disadvantages :
    • Inconveniences with the cable (entanglements, cable damage, etc.)
    • Workspace limited by the length of the cable

​

​

​

L.Peternel@tudelft.nl

13 of 13

FINAL REVIEW MEETING

Luka Peternel

Cognitive Robotics, 3mE

Delft University of Technology

The Netherlands

​

​

ESA Rhizome project

L.Peternel@tudelft.nl