07. MCL in Action

MCL in Action

MCL vs EKF in Action

1- MCL:

Source: Thrun, 2005 - Probabilistic Robotics

Source: Thrun, 2005 - Probabilistic Robotics

At time:

  • t=1, Particles are drawn randomly and uniformly over the entire pose space.
  • t=2, Measurement is updated and an importance weight is assigned to each particle.
  • t=3, Motion is updated and a new particle set with uniform weights and high number of particles around the three most likely places is obtained in resampling.
  • t=4, Measurement assigns non-uniform weight to the particle set.
  • t=5, Motion is updated and a new resampling step is about to start.

2- EKF:

Source: Thrun, 2005 - Probabilistic Robotics

Source: Thrun, 2005 - Probabilistic Robotics

At time:

  • t=1, Initial belief represented by a Gaussian distribution around the first door.
  • t=2, Motion is updated and the new belief is represented by a shifted Gaussian of increased weight.
  • t=3, Measurement is updated and the robot is more certain of its location. The new posterior is represented by a Gaussian with a small variance.
  • t=4, Motion is updated and the uncertainty increases.

Select concepts that are common to both the MCL and EKF algorithms

SOLUTION:
  • Gaussian distribution
  • Motion and measurement stages

Which approach do you prefer?

SOLUTION:
  • MCL