CHILD (Controller for Humanoid Imitation and Live Demonstration): a Whole-Body Humanoid Teleoperation System

Noboru Myers🤖, Obin Kwon🤖, Sankalp Yamsani, Joohyung Kim*
KIMLAB, University of Illinois Urbana-Champaign
🤖 These authors equally contributed to this work. * Denotes corresponding author.
Paper Coming Soon! Video Github

Abstract

Recent advances in teleoperation have demonstrated robots performing complex manipulation tasks. However, existing works rarely support whole-body joint-level teleoperation for humanoid robots, limiting the diversity of tasks that can be accomplished. This work presents Controller for Humanoid Imitation and Live Demonstration (CHILD), a compact reconfigurable teleoperation system that enables joint level control over humanoid robots. CHILD fits within a standard baby carrier, allowing the operator control over all four limbs, and supports both direct joint mapping for fullbody control and loco-manipulation. Adaptive force feedback is incorporated to enhance operator experience and prevent unsafe joint movements. We validate the capabilities of this system by conducting loco-manipulation and full-body control examples on a humanoid robot and multiple dual-arm systems. Lastly, we open-source the design of the hardware promoting accessibility and reproducibility.

Videos

Hardware Design

rl

The hardware system comprises of a compact "torso" designed to fit within a standard baby carrier. The "torso" has seven pluggable mounts which enables the system to be easily reconfigured to match the follower system. The leaders are kinematically scaled leaders corresponding to each limb. Each leader is a kinematically scaled model of the follower, so the joint positions are directly sent to the follower.

For complex tasks that may require multiple operators to fully control all joints or non-humanoid configurations, the CHILD can be attached to a standard monitor stand.

The full bill of materials and 3D printable models are provided on our github repository.

Software

rl

The system control is summarized in the following figure. The joint states from CHILD are mapped one-to-one to the follower. Additionally, two forms of feedback are provided through the leaders, intrinsic and extrinsic feedback. This helps the operator to avoid dangerous singularities and feel large position errors between the leader and follower. For more details on the force feedback, please refer to PAPRLE

Operation

Loco-manipulation operation

Full body joint space operation

When performing loco-manipulation tasks, the on-board walking controller is used instead of controlling the leg joint states directly. In this case, the operator deactivates one arm to use the corresponding leg as a joystick for locomotion commands. While in locomotion mode, the deactivated arm will maintain the previous state, allowing the operator to hold objects while moving.