Grasping Learning System for Irregular Objects Using Offline Reinforcement Learning and Distribution Shift Penalties

This technology utilizes an offline reinforcement learning model to derive grasping poses for irregular objects. It extracts the workspace from offline data collected within the robot's operating environment and applies a penalty to actions with high Q-values that are not present in the valid offline dataset, thereby preventing excessive Q-value predictions outside the offline data distribution and optimizing the grasping success rate.

Real-time online reinforcement learning methods carry a high risk of robot damage during data collection, are time-consuming and costly, and are difficult to implement in real-world field applications due to environmental constraints.

This technology employs an offline reinforcement learning-based model to perform training without real-time interaction. By inferring pixel-wise Q-values and applying a penalty to actions with the maximum Q-value that are not included in the existing dataset (valid offline data), the model is prevented from selecting incorrect optimal actions outside the training data distribution, allowing for the precise derivation of 6-DOF grasping poses. It can be applied to logistics picking, service robots, and manufacturing automation, optimizing grasping success rates without the risk of robot damage during data collection.

‍

‍

Key Features:

• Offline reinforcement learning model training step, where the training unit of the learning server trains an offline reinforcement learning model using valid offline data and infers pixel-wise Q-values
• Penalty application step, where the action with the highest Q-value among those not present in the valid offline data is selected as a representative action, and a penalty is applied only to that representative action
• Maximum Q-value action determination step, where it is verified whether an action is the one with the highest Q-value among those determined to be excluded from the valid offline data
• Valid offline data extraction step, where valid offline data corresponding to the workspace is extracted

‍

This invention was developed with the support of the Artificial Intelligence Convergence Innovation Talent Cultivation program by the Ministry of Science and ICT.

6-DOF Object Grasping Pose Learning System Combining Deep Reinforcement Learning and Gram-Schmidt Orthogonalization

This technology converts spatial information acquired through an input device into point cloud data and applies a deep reinforcement learning model to infer the Q-value (success rate) and rotation vector for each point. Based on the inferred values, it uses Gram-Schmidt orthonormalization to determine the 6-DOF grasping pose.

Conventional supervised learning-based grasping techniques require sophisticated dynamic models, limiting their use for grasping unknown objects without CAD data. Furthermore, the presence of multiple objects often leads to occlusion, which reduces grasping success rates.

This technology uses a deep reinforcement learning model to infer grasping positions and angles directly from point clouds without the need for label generation. In the event of a learning failure, it calibrates the reward function model through inverse reinforcement learning and optimizes the 6-DOF grasping pose using the Gram-Schmidt orthogonalization technique. Applicable to logistics picking, service robots, and manufacturing automation, it significantly increases the grasping success rate for unknown objects even without CAD data.

‍

‍

Key Features:

• Deep reinforcement learning model training phase, which learns from point cloud data and infers the Q-value and rotation vector for each point.
• Work area spatial information extraction phase, where the data generation unit extracts the portion corresponding to the robot device's workspace from the first spatial information to create second spatial information.
• Point cloud data generation phase, which includes a spatial information reception phase where the learning server's data generation unit receives the first spatial information.
• Spatial information acquisition phase, which acquires the first spatial information containing the target object for grasping through an input device.

‍

This invention was developed with the support of the Artificial Intelligence Convergence Innovation Talent Cultivation program by the Ministry of Science and ICT.

Method for Determining the Position and Angle of a Microrobot

This technology is an image processing method that calculates the 3D position and angle of a microrobot by performing thresholding and noise removal on top-view and side-view images of the microrobot collected within a simulation environment, followed by setting a Region of Interest (ROI) and applying edge detection and line detection algorithms.

Challenges include reduced mobility efficiency due to the miniaturization of microrobots, increased control difficulty for real-world human applications, and a lack of precise state recognition technology for pre-testing and simulation.

This technology utilizes camera images to identify the initial position of the microrobot, performs ROI setting and length-based noise filtering, and then extracts the slope and length of line segments through edge and line detection to ultimately derive the microrobot's rotation angles (Yaw, Roll, Pitch) and position. It can be applied to industrial robots and automation systems, improving the control of medical microrobots by providing a method to clearly recognize their position and angle within a simulation environment.

‍

‍

Key Features:

• Includes the step of determining the position and angle of the microrobot in 3D space from an image, where the step of determining the position and angle includes identifying the initial position of the microrobot from the image.
• A step of removing noise that is longer or shorter than the length of the microrobot, based on the length of the microrobot displayed in the image, from the noise generated during the process of isolating the Region of Interest (ROI) from the image.
• A step of performing edge detection at the final position displayed in the image to identify points of discontinuity based on image brightness.
• A step of setting the initial position in the image as a Region of Interest (ROI) for the microrobot.

‍

This invention was developed with support from the Ministry of Trade, Industry and Energy for the development of a microrobotic system for the treatment of chronic total occlusion in myocardial infarction.

Electro-adhesive Pneumatic/Hydraulic Gripper

This technology is a soft robotic gripper mechanism that combines a flexible gripping unit, which expands and contracts via a pneumatic/hydraulic chamber, with an electro-adhesive film that utilizes electrostatic attraction. This increases static friction with objects, allowing for the secure gripping of irregularly shaped items.

Conventional motor-driven robotic hands are limited in their ability to grip irregular objects of various shapes and materials due to their rigid construction, while pneumatic/hydraulic soft robotic grippers often suffer from weak gripping force due to the nature of their flexible materials.

This technology places an electro-adhesive film containing a first electrode in the contact area of the flexible gripping unit to generate electrostatic adhesion. By covering the electrode with a thin film that has lower elongation and higher stiffness than the gripper body, it maintains flexibility while locally enhancing gripping force. Applicable to logistics picking, service robots, and manufacturing automation, it improves both clamping force and durability compared to traditional pneumatic grippers.

‍

‍

Key Features:

• First and second gripping units connected to the main body and driven by the internal pneumatic/hydraulic chamber
• The first gripping unit includes a first electro-adhesive film with a first electrode in the area that contacts the object
• The second gripping unit includes a second electro-adhesive film with a second electrode in the area that contacts the object
• Each electro-adhesive film features a structure where the electrode is encased in a thin film with a lower elongation rate than the gripping unit itself

‍

This invention was developed with support from the Ministry of Trade, Industry and Energy for the development of shape-adaptive electro-adhesive grippers capable of picking irregular multi-objects.

Linkage for Equilibrium Manipulators

This technology features a rod-shaped linkage made of elastic material (rubber or polypropylene) with annular hinge grooves that form joints. It enables 6-DOF position and orientation control through the elastic deformation of the material, eliminating the need for separate joint assemblies.

Conventional manipulator linkages consist of multiple joint assemblies (translational, universal, spherical), which lead to positional errors due to friction and vibration at each joint, as well as structural complexity that hinders precise miniaturization.

By molding the linkage itself from elastic materials like rubber or polypropylene and carving semi-circular hinge grooves at specific locations to induce bending and twisting, this technology eliminates physical friction and enables an ultra-precise, ultra-compact structure. It can be applied to industrial robots and automation systems, improving the precision and control of manipulators by reducing friction and structural limitations.

‍

‍

Key Features:

• Rod-shaped elastic linkage connecting the end effector to the actuator
• One or more joints formed to rotatably support the linkage around its central axis
• Joints provide rotational support in response to actuator movement to change the position and orientation of the end effector
• Structure where the linkage bends and flexes via semi-circular hinge grooves embedded in the joints

Curling Robot

This technology utilizes a three-wheeled robot structure, featuring a pair of front wheels and a single rear wheel, to perform precise turns and changes in direction by calculating the steering angle of the rear wheel based on the kinematic geometry between the front and rear wheels.

It overcomes the limitations of conventional methods that rely on external guide lines or environmental data, enabling active and independent path control using the robot's own kinematic dimensions.

This technology establishes a kinematic triangular model that accounts for the distance between the center of the front wheels and the rear wheel, as well as the distance between the front wheels. By determining the rear wheel steering angle through mathematical formulas, it controls turning maneuvers by fixing the body's center of rotation to one of the front wheels. Applicable to industrial robots and automated systems, it improves the precision and efficiency of a curling robot's movement on ice by controlling its travel direction without external guidance.

‍

‍

Key Features:

• A curling robot for throwing a stone, comprising a body
• A controller that determines the rotation angle of the body, using one of the pair of front wheels as the center of rotation, based on the rotation angle of the rear wheel relative to a virtual line connecting the center of the pair of front wheels and the rear wheel
• A curling robot that rotates the body by the determined rotation angle around the center of rotation while the body is in a stationary state.
• A pair of front wheels installed at the front of the body

‍

This invention was developed with support from the Ministry of Science and ICT for the development of AI curling robot technology capable of establishing game strategies and performing in matches.

Tracking Device

This technology is a mechanism for a mobile tracking device equipped with a positioning module containing multiple positioning nodes, an imaging module, and a sensor module. The device uses a controller to measure the distance to a target node, set a path, and sense obstacles and moving objects to perform path correction, stopping, and re-tracking. In particular, the positioning module calculates the target position through a node configuration of isosceles right triangles and squares, and uses a verification and correction algorithm based on combinations of three nodes.

Existing GPS-based indoor positioning suffers from low accuracy and signal distortion, as well as the inconvenience of having to pre-install positioning nodes throughout the entire area. Furthermore, there is a lack of safety due to the inability to respond to fixed obstacles or sudden moving objects encountered during real-time movement.

This technology utilizes four positioning nodes installed inside the mobile unit to dynamically calculate the distance to a target node. Based on data input from the imaging module and sensor module, the controller controls the real-time path (avoidance, stopping, and restarting). Specifically, it adopts a computational structure that combines three out of the multiple nodes to measure distance, while using the remaining nodes to verify and correct the data. It can be applied to rehabilitation training, gait assistance, and medical/welfare services, improving the accuracy of positioning the target object and enhancing environmental management capabilities by modifying the path based on obstacle detection.

‍

‍

Key Features:

• A tracking device that verifies and corrects the measured distance at least once using three positioning nodes, including the remaining positioning nodes.
• A controller that receives data from the positioning module, imaging module, and sensor module to set the path and control the operation of the drive module.
• A sensor module that senses the front of the main body to detect moving objects between the target object and the main body.
• A tracking device for following a moving target object, comprising a drive module that moves the main body.

‍

This invention was developed with the support of the Ministry of Science and ICT's Real-time Indoor Wide-area Positioning Technology Development project.

Ultra-Precision Compact Parallel Robot Manipulator

This technology features a parallel robot manipulator structure that performs multi-degree-of-freedom position and orientation control. It consists of a frame with top and bottom plates and guide pins, housing piezoelectric motor-based actuators and slide plates that transmit translational motion to an end-effector via linkage units.

Conventional manipulators are difficult to miniaturize due to complex structures where actuators move along axes, and their structural complexity and large volume limit their commercial application in areas such as ultra-precision tasks or medical use.

This technology adopts a parallel kinematic structure where the rotational motion of a screw driven by an actuator (piezoelectric motor) fixed to the bottom plate is converted into translational motion via ball bearings to move the slide plate vertically. The linkage unit connected to the slide plate then drives the end-effector, resulting in a more compact device with improved precision. Applicable to industrial robots and automation systems, it enhances the miniaturization, precision, portability, and control capabilities of parallel robot manipulators.

‍

‍

Key Features:

• Frame unit forming the exterior with top plate, bottom plate, and guide pins
• Multiple independently operated drive units coupled to the frame
• End-effector provided inside the frame unit, with a device mounted on its base
• Linkage units connecting the end-effector to the multiple drive units to control position and orientation

Multi-jointed Robot

This technology features multiple multi-jointed legs coupled to a main body, with friction pads at the base of each leg featuring grooves that provide anisotropic friction. Each leg consists of alternating joint units with a first joint axis and a second joint axis perpendicular to it. By aligning the grooves of the friction pads parallel to the connection direction, the robot mimics snake-like movement and enables multi-jointed walking functionality.

Wheeled robots are limited by road surface conditions, while conventional legged robots often struggle with mobility or maneuverability on rough terrain and suffer from reduced efficiency when utilizing multi-jointed structures.

This technology utilizes a hyper-redundant leg structure that mimics the biological movement of a snake, combined with anisotropic friction pads that induce varying levels of friction based on the direction of ground contact. By forming longitudinal grooves in the friction pads, the design minimizes interference during walking motions and enables terrain-adaptive maneuvering. Applicable to industrial robots and automation systems, this technology enhances robot mobility and adaptability in challenging terrains and environments.

‍

‍

Key Features:

• A leg unit connected to a main body, comprising a first joint unit with a first joint axis and a second joint unit with a second joint axis perpendicular to the first joint axis.
• A leg unit comprising a first joint unit with a first joint axis and a second joint unit with a second joint axis perpendicular to the first joint axis.
• A friction pad unit that provides anisotropic friction to the leg unit.
• A multi-jointed robot where the first direction is parallel to the connection direction in which the first joint unit and the second joint unit are connected.

Collision Detection Sensor

This technology is a collision detection system that utilizes multiple capacitive sensors with varying measurement areas placed on a robot's surface to simultaneously perform wide-range proximity detection and precise position/gesture recognition.

Existing collaborative robots use high-output actuators, which pose a risk of collision accidents when working in close proximity to human operators. Consequently, there is a need for reliable and precise non-contact proximity sensing solutions to prevent such incidents.

This technology features a hybrid array of capacitive sensor groups with different measurement areas (first and second measurement areas) on the robot's link surface, with additional third-area sensors placed at the joints. By leveraging the differences in sensing range and resolution proportional to these areas, the system detects collisions and recognizes user gestures to trigger control actions. Applicable to robotic gripping, precision measurement, and automated equipment, it enhances the safety and reliability of collaborative robots by enabling rapid collision detection and gesture recognition for improved human-robot interaction.

‍

‍

Key Features:

• A collision detection sensor for a robot, comprising a plurality of first capacitive measurement units arranged at regular intervals, and a plurality of second capacitive measurement units, with two or more of the second units disposed between the first units.
• A plurality of second capacitive measurement units disposed on the robot's surface, spaced apart from the first capacitive measurement units, and having a second measurement area different from the first measurement area of the first capacitive measurement units.
• A plurality of second capacitive measurement units having a second measurement area different from the first measurement area of the first capacitive measurement units.
• A plurality of first capacitive measurement units disposed on the surface of the robot.

‍

This invention was developed with support from the Ministry of Trade, Industry and Energy for the development of functional safety implementation technology based on international standards for robots operating in human-contact environments, as well as risk assessment and mitigation technology.

Gripper System

This technology is a gripper system mechanism that physically controls the coefficient of friction between the gripping surface and an object by forming a plurality of micro-channels on the gripping surface and supplying liquid through a fluid pump.

Conventional gripper systems rely solely on normal force control, which can cause deformation when gripping flexible objects. Furthermore, because the coefficient of friction is fixed as a constant value, there are limitations in precisely gripping objects of various materials and shapes.

This technology actively controls the contact area and frictional force by discharging liquid through a plurality of micro-channels formed on the gripping part, and adjusts the supply volume of the fluid pump and the gripping force of the drive unit in real time based on the object's state measured by force sensors. Applicable to logistics picking, service robots, and manufacturing automation, it improves precision control for gripping various materials and objects of different shapes by controlling friction through fluid and deformation control of the gripping part.

‍

‍

Key Features:

• A gripper system for gripping an object, comprising a pair of gripping parts having gripping surfaces with a plurality of micro-channels formed therein, which come into contact with the object through the gripping surfaces.
• A control unit that controls a drive unit to have the pair of gripping parts grip an object, or controls a fluid pump to adjust the amount of liquid supplied to the gripping surfaces.
• A drive unit that operates the pair of gripping parts.
• A fluid pump connected to the plurality of micro-channels to supply liquid.

Microrobot for Drug Delivery

This technology features a mechanical system where a microrobot connected to a base rod uses an external magnetic field to move an internal magnetic linear actuator longitudinally to pressurize and release drugs, or deforms a magnetic absorption member to release the drug.

Conventional balloon catheters are difficult to use in micro-vessels due to the need for radial expansion space, and standalone microrobots are difficult to retrieve after drug release due to blood flow.

This technology utilizes a microrobot fixed to a base rod (catheter/guidewire). It releases drugs by pressurizing them via a magnetically driven linear actuator or by deforming a magnetic absorption member, while the base rod allows for precise positioning and retrieval of the robot. Applicable to surgical robots, interventional systems, and medical automation, it improves drug delivery to small blood vessels and enhances the safety of drug administration.

‍

‍

Key Features:

• The drug reservoir is a linear actuator that moves along the longitudinal axis of the base rod in response to an external magnetic field.
• In a microrobot mounted on a base rod, a connection part is used to secure the robot to the base rod.
• A connection part that secures the robot to the base rod.
• A microrobot where the drug is pressurized and released by a longitudinally moving linear actuator.

‍

This invention was developed with support from the Ministry of Trade, Industry and Energy for the development of a micro-medical robot system for the treatment of chronic total occlusion in myocardial infarction.

Robot Joint Device

This technology is a joint structure that implements 6-degree-of-freedom motion between a first base and a second base based on a parallel mechanism, and calculates rotation angles and torque through a sensor unit that includes a rotation angle measurement component and an elastic component disposed on the rotation axis.

Conventional torque measurement methods require separate torque sensors, which complicates the device structure, increases manufacturing costs, and reduces the overall price competitiveness of the robot due to the use of expensive components.

This technology simplifies the structure by coaxially arranging a torsion spring (elastic component) and a rotation angle measurement component on the link rotation axis. By calculating the measured rotation angle and a predefined elastic coefficient in the control unit, it precisely measures joint torque without the need for a separate torque sensor. It can be applied to robot gripping, precision measurement, and automation equipment, providing a compact and cost-effective robot knuckle device for measuring rotation angles, linear displacement, and power or torque, thereby improving the accuracy and cost-efficiency of robot systems.

‍

‍

Key Features:

• A robot joint device, wherein a plurality of sensor units each comprise a rotation angle measurement component for measuring the rotation angle of a rotor with respect to a rotation axis, and an elastic component for providing elastic force to a connecting member with respect to the rotation axis.
• A plurality of connecting members connecting the first base and the second base to enable 6-degree-of-freedom rotational movement of the second base relative to the first base
• A plurality of sensor units connected to each of the plurality of connecting members to measure the rotation angle of the rotor with respect to the rotation axis and the torque acting on the rotation axis
• A second base disposed spaced apart from the first base

Wearable Device for Lumbar Muscle Support

This technology provides assistive force by stacking multiple unit modules to track the multi-degree-of-freedom movements of the human spine, such as flexion, extension, and lateral bending, while controlling the tension of drive and auxiliary wires. By combining ball/universal joints in the articulation sections with the restorative force of elastic components, it achieves variable stiffness and assistive force tailored to the wearer's spinal movement.

Existing wearable muscle support devices often fail to fully accommodate the complex degrees of freedom of the spine (extension, flexion, lateral bending, rotation, etc.), resulting in limited support ranges and causing discomfort or restricted movement for the wearer.

This technology utilizes drive wires and left/right auxiliary wires that pass through multiple unit modules arranged along the spine, with a drive module that variably controls the tension of each wire. It provides lateral bending support through elastic components and optimizes muscle assistance for the wearer's movements by measuring and providing feedback on wire tension via pulley and spring encoders. Applicable to rehabilitation training, gait assistance, and medical/welfare services, it provides a wearable device that offers high-degree-of-freedom muscle support, prevents lower back injuries, and reduces lumbar load, thereby improving comfort and transmission characteristics.

‍

‍

Key Features:

• Auxiliary wires passing through the multiple unit modules along the spinal direction on the left and right sides of the modules
• A drive module that operates the drive wire to generate and adjust the tension of the drive wire
• A spinal module comprising a plurality of unit modules arranged along the spinal direction
• A drive wire passing through the center of the plurality of unit modules along the spinal direction

‍

This invention was developed with support from the Ministry of Trade, Industry and Energy for the development of international standard-based functional safety implementation technology and risk assessment/reduction technology for robots operating in human-contact environments.

Magneto-Acoustic System for Microrobot Steering

This technology utilizes an acoustic generator installed on an external base to create pressure differences from standing waves within a fluid medium, focusing multiple microrobots into a specific point to form a swarm. A magnetic field generator then creates a field to steer and move the swarm to a target location.

Using a single microrobot makes efficient drug delivery difficult due to limited storage capacity, and moving individual robots is time-consuming and costly.

This technology is a magneto-acoustic steering system and method that forms a microrobot swarm by applying sound waves to the fluid medium inside an object using multiple acoustic elements placed on a base, and then controls a magnetic field generator to stably guide the swarm to a target point. It can be applied to industrial robots and automated systems, improving drug delivery capacity and the steering of multiple microrobots.

‍

‍

Key Features:

• A magnetic field generator installed on the base that creates a magnetic field within the object.
• An acoustic generator installed on the base that transmits sound waves into the object.
• A magneto-acoustic system for microrobot steering that uses the pressure difference at the nodes of superimposed standing sound waves to focus multiple microrobots.
• Multiple microrobots injected into the object.

‍

This invention was developed with support from the Ministry of Science and ICT for the Intelligent Microrobot-based Body-on-a-Chip for Precision Medicine project.