Look naturally.
HEAD → HMD
Head motion guides the robot’s gaze. Stereo vision brings the robot’s view back to you.
A Hybrid Interface for Humanoid Teleoperation
with Multimodal Sensing and Exoskeleton Control
Department of Robotics, Kwangwoon University / Seoul, Republic of Korea
* Corresponding author
01 / THE INTERFACE
Different parts of the body move differently. HITMEN assigns each one an interface that fits—so looking, reaching, leaning, and grasping become part of one continuous action.
HEAD → HMD
Head motion guides the robot’s gaze. Stereo vision brings the robot’s view back to you.
CHEST → VR CONTROLLERS
Chest-mounted controllers capture how you turn and lean, independently of your head.
ARMS → LEADER ARMS
Shoulder-mounted leader arms map your joint motion directly to the robot’s arms.
HANDS → VR + RGB CAMERAS
Palm-facing wrist cameras keep fingers in view when headset hand tracking struggles.
FULL SYSTEM VIDEO
Explore the interface and see the full system in action.
02 / ANTICIPATING OCCLUSION
A hand turns. An arm crosses the chest. The headset’s view changes—and so does the quality of its tracking.
HITMEN learns from recent motion to anticipate the next second of hand and controller tracking reliability. When VR hand tracking becomes unreliable, finger control shifts toward the wrist-camera estimate.
Watch the estimated reliability change as hand poses affect the headset’s view.
The simulated robot hand stays responsive even when the VR tracking state reports a failure.
03 / INDEPENDENT HEAD & TORSO
Your gaze should not decide every movement of your body.
HITMEN gives the head and chest separate targets. The neck follows the HMD; the waist follows the chest controllers. That separation lets you change your viewpoint while keeping your torso where the task needs it.
Head and torso rotation follow their own motion targets.
Coordinate the waist and neck while preserving separate head and chest targets.
04 / HITMEN IN ACTION
From lining up a precise insertion to moving a hand beyond the headset’s tracking coverage, the interface makes more of the robot’s upper body available to the operator.
Guide, align, and insert a peg into a hole in simulation.
Palm-facing cameras support finger control beyond the headset’s hand-tracking range.
05 / SIDE-BY-SIDE BENCHMARKS
Four selected manipulation tasks, shown across three interfaces. Watch how each setup handles coordination, occlusion, and delicate contact.
“Play all” starts the three clips together. Clips retain their original timing and are not aligned by task phase. Swipe each row to compare all three interfaces.