Department News
Autonomous Exploration in Unknown Urban Environments for Unmanned Aerial Vehicles
Seminar Date
2005-09-09
Author
빈종훈
Date
2005-09-06
Views
1869
1. 제 목 : Autonomous Exploration in Unknown Urban Environments for Unmanned Aerial Vehicles
2. 연 사 : Dr. Hyunchul Shim
3. 일 시 : 2005년 9월 9일 (금) 오전 11:00 - 12:00
4. 장 소 : 서울대 301동 117호
5. 내 용 :
As UAVs their way into more demanding applications such as low-altitude ground support or urban operations, they are expected to fly autonomously without colliding into obstacles. The capability to avoid obstacles has not been recognized as a major concern to conventional UAVs, which are typically operated at higher altitudes and closely monitored by human operators. However, as UAVs are required to operate in cluttered or dynamically changing environments without human support, the autonomous exploration capability has been recognized as a crucial component technology for future UAVs.
Exploring autonomously in an unknown environment requires map building and online trajectory replanning for collision-free navigation. Many existing algorithms are limited to two-dimensional cases, computationally expensive or not suitable for real-time applications. UAVs also pose more challenges than the ground robots do in the development, implementation, and operation: they typically fly at a much faster speed than the ground robots, and therefore requires more accurate decision making in much shorter time.
We also consider a number of obstacle sensing methodologies. The information about obstacles in the flight path can be made available to the path planner by a pre-programmed map or a dynamically built map. Whereas the former approach does not suffer from any sensor-induced errors, the map itself can be inaccurate or outdated. Therefore, we favor dynamic map building using onboard sensors. Laser range finders, which measure the distance from the sensor to objects around, have been found very attractive due to its accuracy and long detection range.
Model predictive control has been found attractive for addressing control problems with many dynamic and operational constraints. The online optimization with finite horizon enables a control system more responsive to the changes in the system dynamics and the surroundings. Further, it has been explored and proven effective that a variety of performance goals, in addition to the feedback stabilization, can be incorporated into the cost function. Particularly, it is shown that MPC is capable of obstacle avoidance using a cost function that penalizes the proximity to the nearest obstacle.
In this talk, an autonomous exploration algorithm suitable for, but not limited to, urban navigation by combining the MPC-based obstacle avoidance with local obstacle map building using onboard laser scanning is presented. Starting from the given trajectory, the MPC layer solves for a collision-free trajectory by the real-time gradient-search based optimization. The proposed framework is validated in simulations, and then tested in a series of experiments using a simulated urban environment as will be shown in the talk.
6. 약 력 :
2005.3 Present Project Manager and principal development engineer, Berkeley UAV Research
Group, University of California, Berkeley, USA
2001.6 2005.2 Staff Engineer, Maxtor Corporation, USA
2000.12 2001.6 Specialist, Berkeley UAV Research Group, University of California, Berkeley, USA
2000.12 Ph.D., Dept. of Mechanical Engineering, University of California, Berkeley, USA
1993. 2 M.A., Dept. of Mechanical Design and Production Engineering, Seoul National University
1991.2 B.A., Dept. of Mechanical Design and Production Engineering, Seoul National University
7. 문 의 : 기계항공공학부 기 창 돈 (☏ 880-1912)
2. 연 사 : Dr. Hyunchul Shim
3. 일 시 : 2005년 9월 9일 (금) 오전 11:00 - 12:00
4. 장 소 : 서울대 301동 117호
5. 내 용 :
As UAVs their way into more demanding applications such as low-altitude ground support or urban operations, they are expected to fly autonomously without colliding into obstacles. The capability to avoid obstacles has not been recognized as a major concern to conventional UAVs, which are typically operated at higher altitudes and closely monitored by human operators. However, as UAVs are required to operate in cluttered or dynamically changing environments without human support, the autonomous exploration capability has been recognized as a crucial component technology for future UAVs.
Exploring autonomously in an unknown environment requires map building and online trajectory replanning for collision-free navigation. Many existing algorithms are limited to two-dimensional cases, computationally expensive or not suitable for real-time applications. UAVs also pose more challenges than the ground robots do in the development, implementation, and operation: they typically fly at a much faster speed than the ground robots, and therefore requires more accurate decision making in much shorter time.
We also consider a number of obstacle sensing methodologies. The information about obstacles in the flight path can be made available to the path planner by a pre-programmed map or a dynamically built map. Whereas the former approach does not suffer from any sensor-induced errors, the map itself can be inaccurate or outdated. Therefore, we favor dynamic map building using onboard sensors. Laser range finders, which measure the distance from the sensor to objects around, have been found very attractive due to its accuracy and long detection range.
Model predictive control has been found attractive for addressing control problems with many dynamic and operational constraints. The online optimization with finite horizon enables a control system more responsive to the changes in the system dynamics and the surroundings. Further, it has been explored and proven effective that a variety of performance goals, in addition to the feedback stabilization, can be incorporated into the cost function. Particularly, it is shown that MPC is capable of obstacle avoidance using a cost function that penalizes the proximity to the nearest obstacle.
In this talk, an autonomous exploration algorithm suitable for, but not limited to, urban navigation by combining the MPC-based obstacle avoidance with local obstacle map building using onboard laser scanning is presented. Starting from the given trajectory, the MPC layer solves for a collision-free trajectory by the real-time gradient-search based optimization. The proposed framework is validated in simulations, and then tested in a series of experiments using a simulated urban environment as will be shown in the talk.
6. 약 력 :
2005.3 Present Project Manager and principal development engineer, Berkeley UAV Research
Group, University of California, Berkeley, USA
2001.6 2005.2 Staff Engineer, Maxtor Corporation, USA
2000.12 2001.6 Specialist, Berkeley UAV Research Group, University of California, Berkeley, USA
2000.12 Ph.D., Dept. of Mechanical Engineering, University of California, Berkeley, USA
1993. 2 M.A., Dept. of Mechanical Design and Production Engineering, Seoul National University
1991.2 B.A., Dept. of Mechanical Design and Production Engineering, Seoul National University
7. 문 의 : 기계항공공학부 기 창 돈 (☏ 880-1912)
