Full-Body Motion Planning
for Locomotion on Uneven Terrain





Presented by
Henrique Ferrolho

Supervision
Rosaldo Rossetti, University of Porto
Vladimir Ivan, University of Edinburgh

Mars











The ultimate goal

Agenda


  1. Introduction
  2. Motivation
  3. Problem
  4. Related Work
  5. Goals
  6. Methodology
  7. Work Plan

Introduction

Humanoid Robots

Valkyrie, Atlas, and Robonaut.

Introduction

Humanoid Robots

Characteristics:

  • Floating base
  • Safe human interaction
  • Complex environments
  • Dynamic environments
  • Balance constraints


Introduction

Humanoid Robots


Challenges:

(Some of them...)
  • Automation
  • Bipedal balance
  • External factors
  • Many degrees-of-freedom

Motivation


Humanoid robots capable of autonomously planning and reaching difficult grasping goals in more realistic, i.e. more complex and dynamic environments.

Problem

State of the Art
Darpa Robotics Challenge. [PM13]

Problem


  • Engineering (Informatics)
    • Robotics and AI
      • End-Pose Planning (Whole-Body)
        • on Uneven Terrains

Problem

End-Pose Planning

Essential problem in humanoid applications:

  • Valid stance locations
  • Collision-free reaching configurations


Non-trivial in complex environments.
Serves as the input to walking and motion planners.

Problem

End-Pose Planning

 
Quadratic
Programming (QP)


  • Hard constraints
  • Slow
Jacobian Inverse Kinematics (IK)

  • Soft constraints
  • Fast

  • Obstacle avoidance is too expensive
  • Solvers get stuck in local minima

Related Work

iDRM

  • Offline stage: iDRM construction
  • Online stage: valid end-pose selection

Related Work

 
Offline stage: sampling.

Related Work

 
Given a grasping goal, where can the robot stand?
Quadrant of an iDRM. [YIL+16] Voxels are colored according to their amount of
collision-free states (green = high, red = low).

Related Work

iDRM: System Overview

  1. End-pose planning
  2. Footstep planning
    • Walking execution
  3. Motion planning (Whole-Body)
    • Motion execution

Related Work

1. End-pose planning

[YIL+16]

Related Work

2. Footstep planning

[YIL+16]

Related Work

2.1. Walking execution

[YIL+16]

Related Work

3. Motion planning

[YIL+16]

Related Work


Example of an unnatural-looking pose.
Open problems:

  • Inclined terrains
    • Combinatorial overhead
  • Unnatural-looking poses
  • Bimanual manipulation

Goals


Enable humanoid robots to autonomously
plan complex end-poses on uneven terrains.


  • Extend dataset
    • Split dataset in 2: upper-body and lower-body
  • Improve samples
    • Pseudorandom sampling with constraints for human-like poses

Methodology


  • Explore current problems
  • Design an approach
  • Implement solution
  • Simulate
  • Test with robot
  • Evaluate results

Methodology

Part of the testbed.

Methodology

Simulation Construction Set (left) and Director (right).

Work Plan

2017

Thank you


Special thanks to Wolfgang Merkt and Yiming Yang for
putting up with me and reviewing this presentation.

References


[ES14] Elbanhawi, Mohamed, and Milan Simic. "Sampling-based robot motion planning: A review." IEEE Access 2 (2014): 56-77.
[NASA17] NASA. "Mars planet facts news & images." http://mars.jpl.nasa.gov (accessed January 29, 2017).
[PM13] Pratt, Gill, and Justin Manzo. "The darpa robotics challenge [competitions]." IEEE Robotics & Automation Magazine 20, no. 2 (2013): 10-12.
[YIL+16] Yang, Yiming, Vladimir Ivan, Zhibin Li, Maurice Fallon, and Sethu Vijayakumar. "iDRM: Humanoid motion planning with realtime end-pose selection in complex environments." In Humanoid Robots (Humanoids), 2016 IEEE-RAS 16th International Conference on, pp. 271-278. IEEE, 2016.
[YIMV16] Yang, Yiming, Vladimir Ivan, Wolfgang Merkt, and Sethu Vijayakumar. "Scaling Sampling-based Motion Planning to Humanoid Robots." In ROBIO. IEEE, 2016.