Towards Designing Benchmarks for Humanoid Space Robots: Northeastern’s
NASA Valkyrie Dataset
Abstract
Designing benchmarks for algorithms developed to run on physical robot
hardware remains to be a challenge. Towards achieving benchmarks for
space humanoid robots, we present Northeastern’s humanoid robot dataset
containing physical sensor data from NASA’s Valkyrie (R5), including
robot pose estimate, joint angles and velocities, center of pressure,
center of mass, ground reaction wrenches, and motion capture ground
truth pose. The dataset includes various mobility and manipulation tasks
as atomic robot behaviors including walking and reaching motions.
Inspired by the NASA Space Robotics Challenge, the dataset is intended
for use by the community that wishes to conduct humanoid robot research
without direct access to a hardware platform. In addition, it will
enable comparative studies in terms of hardware designs, as well as task
and motion planning methods. The dataset will provide the humanoid
robotics research community with a resource not only to bridge the gap
between simulation-based and experimental algorithm validation but also
to design task-level benchmarks for humanoid space robots. This paper
describes the robot hardware, software, data collection process,
post-processing steps, and structure of data for Northeastern’s NASA
Valkyrie dataset.