• Welcome Message
    • KUME
    • member
    • direction
    • International Students
    • Undergraduate
    • Graduate
    • Research Area
    • Laboratory
    • Notice
    • Q&A


  • Home
  • about us
  • member
  • professors
professors emeritus
Jaebok Song
Jaebok Song | Professor
School of Mechanical Engineering, Korea University, Anam-dong, Seongbuk-gu, Seoul, 136-713, Korea (South)
Feb. 1983 B.S. in Mechanical Engineering, Seoul National University, Seoul, Korea
Feb. 1985 M.S. in Mechanical Design and Production Engineering, Seoul National University, Seoul, Korea
Jan. 1992 Ph.D. in Mechanical Engineering, MIT, USA
Intelligent Robotics Lab.
Total Solution to Collision Safety between Humans and Robots
Physical human-robot interaction becomes increasingly important as service robots work in the human environments and industrial robots operate in collaboration with workers. To deal with collision safety, three-step safety strategies were proposed in our lab; as step 1, collision prediction and safe motion (prediction of collision between human and robots using vision camera, and path regeneration with considering collision safety and efficient performance); as step 2, collision detection and reaction (detection of collision by sensor and reaction to reduce or eliminate collision force when step 1 fails; and as step 3, collision absorption (collision absorption using a passive device made of only mechanical components when steps 1 and 2 fail.
Safety Mechanisms for Collision Absorption
Two approaches are usually used to achieve collision safety related to physical human-robot interaction; an active approach based on sensors and actuator control, and a passive approach with only mechanical devices. Since sensors can be affected by disturbance from the environment and the subsequent motor response is slow, the active approach has limitations in achieving the high-level of collision safety. To deal with these problems, a safe joint mechanism (SJM) and a safe link mechanism (SLM) were suggested in our lab. The SJM and SLM were designed such that their stiffness is kept high in normal operation for positioning accuracy but becomes low in response to a high external force for collision safety. These mechanisms, composed of only mechanical components such as springs and a 4-bar linkage system, can provide fast response speed and high reliability.
Collision Detection and Response
A robot must be able to detect a collision between a robot and an environment (including a human) during its motion, and to properly respond to this collision. The collision detection and response algorithms based on joint torque sensors were developed in our lab. To compensate for the motion effects, disturbance and sensor noise, the detection algorithm uses a residual observer based on a generalized momentum. When a collision is detected, the robot executes an emergency stop or reflex motion to minimize the impact delivered to a human.
Variable Stiffness Actuation
Either position control or force control is available for one degree of freedom of a robot joint unlike human joints that can execute position and force control simultaneously. To deal with this limitation of a robot joint, the serial and parallel types of dual actuator units were developed in our lab. A serial-type dual actuator unit (SDAU) composed of two actuators, connected in series, can control the position and stiffness simultaneously for the same joint by exploiting the features of a planetary gear train. Without an additional force/torque sensor, the contact force can be controlled by using the position information of the two actuators, and the stiffness and shape of the environment can be estimated as well. A parallel-type dual actuator unit (PDAU), using two actuators connected antagonistically via the nonlinear spring mechanism, also provides the simultaneous control of position and stiffness at the same joint. Moreover, the PDAU can immediately change a joint stiffness to a very low value when an external force greater than a predetermined threshold occurs, so that it offers collision safety without an expensive joint torque sensor.
Guard Robots for Social Security
A guard robot can be used for reconnaissance in dangerous areas such as at the scene of a fire or a crime. To cope with different situations, a guard robot should be small-sized and lightweight to increase its portability, as well as it should be able to overcome relatively high obstacles. The Guardian is a jumping robot based on a conical spring, and it can reach a higher place more quickly than other locomotion methods. Using the independent control of the jump angle and jump height, the robot can autonomously climb a stair. On the other hand, a multi-active crawler robot (MACbot) is a tracked robot, which can provide the multiple functions with a minimal number of motors. The MACbot can travel with four driving tracks and sometimes these tracks can be rotated to overcome obstacles. This robot is also capable of climbing stairs.
Autonomous Navigation in Outdoor Environments
Outdoor navigation techniques can be applied to various fields including general service robots, unmanned ground vehicles, and military robots. For autonomous outdoor navigation, mapping and localization in outdoor environments were developed in our lab. Our outdoor mobile platform is equipped with various sensors such as a laser scanner, a GPS/INS module, a stereo camera, and wheel encoders. Outdoor environments are modeled by integrating various sensor data and the robot pose is accurately estimated using both the sensor data and the world model for the outdoor environment. Sensor fusion and reliable matching methods were developed in our lab to improve the performance of outdoor navigation.
Autonomous Navigation in Large and Dynamic Indoor Environments
Public service robots, such as guide robots, require safe and reliable navigation in public places where many people and obstacles exist. To this end, autonomous indoor navigation algorithm in large and dynamic environments was developed in our lab. It uses a laser scanner which can accurately scan a large area and sonar sensors which detect the objects unseen by the laser scanner. The robot can accurately estimate its pose in large indoor environments, and can get to the goal while avoiding various types of obstacles. Furthermore, the robot can build a map of the environment even though the environment is completely unknown.
Low-cost sensor-based Indoor Navigation
SLAM (simultaneous localization and mapping) for inexpensive robots is one of the most challenging issues in mobile robotics. Since the robot is equipped with low-cost sensors, it is important to handle inaccurate sensor information to accurately estimate the robot pose. In this research, an upward monocular camera is used as a main sensor, and a gyroscope and an accelerometer are used to cope with various situations such as wheel slippage and kidnapping. The EKF (extended Kalman filter) scheme is adopted to perform SLAM and to integrate the observations from each sensor. The system uses only the natural landmarks such as corners, lamps, and doors without the aid of any artificial landmark. This low-cost SLAM system is applicable to cleaning robots, security robots, guide robots, and so on.