Source-linked AI summary
Warping the Workspace: Expanding Visual Access with Adjustable Reach Mapping for Humanoid Teleoperation
Keigo Minamida, Koya Dendo, Yuto Nakamura, Jun Rekimoto
TL;DR
First-person humanoid teleoperation ties access to side and rear targets to operator rotation, while panoramic warping alone can break the correspondence between displayed and commanded directions. The paper couples a panoramic azimuth warp with inverse upper-limb yaw remapping and evaluates two implementations. In the sample, GZ had the largest completion count, whereas F+C received the most preference votes.
Problem
Panoramic warping can bring workspace targets into the frontal view, but visual remapping alone can leave displayed target position mismatched with commanded arm direction.
Source: §1 · p. 2Method
Results
23/24 completions for GZ was the largest count, whereas F+C received 6/8 preference votes in this sample.
Source: §3 · p. 2Takeaways & Limitations
Completion and preference favored different mappings in this sample, so the conclusion treats them as separate design criteria.
Source: §4 · p. 2
Abstract
from arXiv · showhide
First-person humanoid teleoperation ties access to side and rear targets to operator rotation. We present coupled visual--motor remapping, pairing a panoramic azimuth warp with its inverse upper-limb yaw transform. A working prototype implements Global Zoom (GZ) and Focus+Context (F+C). In a within-subject study (N=8); 72 trials), GZ had the largest completion count (23/24; Fixed: 20/24; F+C: 21/24), whereas F+C was preferred by 6/8 participants. Therefore, completion and preference favored different mappings in this sample.
1 Introduction
First-person teleoperation preserves an intuitive view–motion correspondence but ties access to side and rear targets to large operator rotations. The paper proposes coupling panoramic view warping with inverse reach remapping, implemented in a humanoid prototype and compared in an exploratory study.
1 Introduction
- First-person systems commonly preserve one-to-one correspondence among operator view, body motion, and robot motion, while side and rear access can require large head and torso rotations.Widening the view can expose more workspace but reduces local visual detail.Source: §1 · pp. 1–2
- Panoramic warping can redistribute a 360° workspace into the frontal field of view, but visual remapping alone can separate a target’s displayed position from its commanded arm direction.The paper treats visual-space allocation as part of the action mapping rather than as display-only zoom.Source: §1 · p. 2
- Prior approaches include mirrored body motion, camera steering, auxiliary views, and modified control mappings; visual-detail redistribution methods include fisheye, peripheral compression, and zoom.The cited focus+context precedent addressed mobile-robot navigation, whereas this approach targets humanoid manipulation with coupled view and reach remapping.Source: §1 · p. 2
- Coupled visual–motor remapping pairs a panoramic view warp with inverse upper-limb yaw remapping so displayed and commanded reach directions remain aligned.The prototype implements uniform Global Zoom and non-uniform Focus+Context, comparing direction-specific completion and overall preference.Source: §1 · p. 2
2 System
The system maps robot-relative azimuth into display azimuth and applies the inverse mapping to wrist control, aligning rendered and commanded reach directions. A 360° camera view is streamed to a headset, which returns tracking and viewport state to the host.
2 System
- Rendering uses the azimuth warp, while control applies its inverse to robot reach, geometrically matching displayed and commanded directions.After pose conversion to a common robot-aligned frame, the mapping takes robot-relative azimuth as input; robot reach azimuth is the control output.Source: §2 · p. 2
- A RICOH THETA X on a Unitree G1 streams a monocular 360° view to an Apple Vision Pro; the headset returns tracking and viewport state for wrist-pose remapping before inverse kinematics.The host performs the remapping between headset tracking and robot control.Source: §2 · p. 2
- Fixed uses identity mappings, Global Zoom uniformly maps 360° into 180° with yaw gain 2.0, and Focus+Context expands the central ±15° region by 1.5× while compressing the periphery.Focus+Context uses the corresponding inverse control warp.Source: §2 · p. 2
3 Evaluation
In this sample, Global Zoom yielded the most completed trials, while Focus+Context received the most preference votes. Eight participants completed 72 trials in a counterbalanced, fixed-base upper-limb-control evaluation.
3 Evaluation
- Eight participants performed one trial per condition and target direction in a counterbalanced 72-trial evaluation using fixed-base, upper-limb control.The task involved front, side, and rear targets, as depicted in Figure 4.Source: §3 · p. 2
- GZ produced the largest completion count, 23/24, whereas F+C received six of eight preference votes; therefore, completion and preference favored different mappings in this sample.Source: §3 · p. 2
4 Conclusion
The conclusion frames view allocation as part of action mapping and distinguishes task completion from user preference as design criteria.
4 Conclusion
- In this sample, GZ yielded the most completed trials while F+C received the most preference votes, motivating separate evaluation of task completion and user preference.The paper concludes that coupled remapping makes view allocation part of the action mapping.Source: §4 · p. 2