Geoff Fink presents Computer Vision-Based Motion Control and State Estimation for Mobile Robotics
On 2021-03-16 12:00
at https://feectu.zoom.us/j/96560335215
Abstract: Mobile robots are increasingly being used for a wide range of both
indoor and outdoor applications such as search and rescue, surveillance, and
infrastructure inspection. However, most are still being remotely piloted. In
this talk I will discuss three of our recent research efforts towards improving
the autonomy of mobile robots. First, I will discuss state estimation. State
estimators for individual robots are often general and many are even designed as
an after thought whereas the preference is given to the mechanical design and
control of the robot. Furthermore, the difficulty of state estimation is often
underestimated. However, stable state estimates are essential for most control
algorithms. We developed a low level state estimator for quadrupedal robots that
includes attitude, odometry, ground reaction forces, and contact detection. The
second topic is dynamic visual servoing (VS). A typical visual servo control
consists of a two loop architecture where the outer loop uses a vision sensor to
provide a reference velocity to the inner loop and the inner loop regulates the
velocity of the robot by providing force and torque commands. Then the vehicle
tracks the reference velocity using a kinematic model. With high speed tasks and
underactuated systems it is important to include the dynamics of the vehicle. We
refer to VS that directly accounts for vehicle dynamics as dynamic VS. The last
topic is visual inertial simultaneous localization and mapping (SLAM). Our work
uses the output of an existing monocular visual SLAM system which provides a
scaled position measurement. Using an observer design, inertial measurements are
combined with the visual SLAM output to estimate the vehicle position and linear
velocity. We consider the observability of this visual inertial SLAM problem and
propose an observer design based on a change of coordinates which transforms the
system into a LTV form. Our approach does not require an approximate
linearisation of the model equations.
Geoff Fink is a candidate for the Robotics tenure-track position at the Dept. of
Cybernetics.
indoor and outdoor applications such as search and rescue, surveillance, and
infrastructure inspection. However, most are still being remotely piloted. In
this talk I will discuss three of our recent research efforts towards improving
the autonomy of mobile robots. First, I will discuss state estimation. State
estimators for individual robots are often general and many are even designed as
an after thought whereas the preference is given to the mechanical design and
control of the robot. Furthermore, the difficulty of state estimation is often
underestimated. However, stable state estimates are essential for most control
algorithms. We developed a low level state estimator for quadrupedal robots that
includes attitude, odometry, ground reaction forces, and contact detection. The
second topic is dynamic visual servoing (VS). A typical visual servo control
consists of a two loop architecture where the outer loop uses a vision sensor to
provide a reference velocity to the inner loop and the inner loop regulates the
velocity of the robot by providing force and torque commands. Then the vehicle
tracks the reference velocity using a kinematic model. With high speed tasks and
underactuated systems it is important to include the dynamics of the vehicle. We
refer to VS that directly accounts for vehicle dynamics as dynamic VS. The last
topic is visual inertial simultaneous localization and mapping (SLAM). Our work
uses the output of an existing monocular visual SLAM system which provides a
scaled position measurement. Using an observer design, inertial measurements are
combined with the visual SLAM output to estimate the vehicle position and linear
velocity. We consider the observability of this visual inertial SLAM problem and
propose an observer design based on a change of coordinates which transforms the
system into a LTV form. Our approach does not require an approximate
linearisation of the model equations.
Geoff Fink is a candidate for the Robotics tenure-track position at the Dept. of
Cybernetics.