LAVER LABS

NAVER LABS has been developing a new haptic device that can capture the natural motion of users while providing force feedback. This enables bilateral teleoperation when coupled with a robot such as AMBIDEX. Users can control and feel what the robot feels, as if the haptic device and the robot are physically connected. In addition to seamlessly connecting the user and robot, any work performed by the user is captured as training data. We aim to use the bilateral haptic device as an effective tool to gather high-quality training data, to advance the intelligence of the dual-arm robot—eventually creating an autonomous robot that can perform tasks on its own. 

▶︎Related article: Haptic device - A medium to enable robots to learn human physical intelligence

The haptic device, which instantly transmits force and movement to the user, is an essential medium to manipulate and touch objects connected to the digital world. The applications are endless, including remote surgery robots, remotely performing dangerous tasks, and various XR services. However, many haptic devices to date can only transmit small amounts of force or have a narrow workspace and many limitations. These limitations have prevented haptic devices from capturing the full range of motions and strategies that humans can carry out.

The newest haptic device is human-scale, which can enable teleoperated control of the dual-arm robot in a human-like manner. Recent advancements at NAVER LABS have produced dramatic improvements to the haptic device’s workspace, force control bandwidth, and transparency, all of which enable the user and robot to perform more complicated tasks.

The workspace refers to the size of a space where the operator can transmit motions through the haptic device. Since haptic devices are held and operated by a person, a larger workspace allows more dynamic movement resulting in the acquisition of training data containing various motions.

Force control bandwidth and transparency mean how quickly the robot can react according to the movement of the operator and how effectively the user can feel the force felt by the robot. Several factors hinder this bandwidth and transparency. The inertia of the haptic device itself and large friction in the motor are factors obstructing the operator’s movement.

Addressing these two simultaneously is very difficult from a technological standpoint because multiple factors are complexly intertwined. In general, if we expand the workspace and range of force transmitted, the size of the haptic device tends to increase. As the device has more inertia, force transparency and force control bandwidth decrease. To overcome this, we applied an optimal design that reflects humans’ work convenience and developed an actuator that can produce the required performance. With Haptic Device 2.0, we successfully maximized the operator's workspace while securing sufficient force control bandwidth and innovatively increased force transparency. 

The videos below show the latest demonstrations using Haptic Device 2.0.  Compared to the previous version, the task performance for complex and detailed movements has significantly improved.

 

Haptic Device 2.0 Demo: IKEA chair assembly

A robot must be able to put each part in the right position and use the appropriate force to insert or tighten the particular piece, to assemble a chair successfully. Traditional robot task learning approaches do not work well in tasks where both ‘position’ and ‘force’ must be considered. With the improved haptic device, we used various human-like movements such as the dual-arm robot pressing down on parts with its arms or leaning against a piece to support it, to successfully complete the difficult task of assembling a chair without any special additional devices. 

 

Haptic Device 2.0 Demo: Table tennis

The table tennis demo below demonstrates that the upgraded haptic device can also perform complex tasks involving quick motions. The racket must swiftly be positioned in the right place in line with the incoming ball. At the moment of impact (when hitting the ball), it must move with sufficient speed and in the correct direction. The table tennis demo would not have succeeded if the haptic device had ever so slightly more inertia or if there was a delay in transmitting the size and direction of the force from the human’s movement to the robot.