• Cyrill Stachniss

• MSR Part 1 Videos
• MSR Part 2 Videos
• MSR Part 2 Slides

Localization - Estimating robot's pose

Mapping - Task of modelling the environment

Where am I --><-- What does the world look like

Full SLAM estimates the entire navigation path whereas, Online SLAM seeks to recover only the most recent pose

1. State estimation - What is the state of the world? Typically, Sensor data required to build the map of the environment.

   • Estimating Semantics - Understanding what we see
   • Estimating Geometry - Understanding what the world looks like

   Both can be fused.

2. Action Selection - Which action should state execute? Typically, physical navigation

These influence each other. It's a chicken-and-egg problem.

Given

1. The robot's controls
2. Sensor Observations

Wanted

1. Map of the environment m
2. Path of the robot

We estimate the probability distribution

• Uncertainty both in robot motion and observations
• Use of probability theory to explicitly represent the uncertainty

1. Robot path and map both are unknown
2. Map and pose estimates are correlated
3. Mapping between observations and the map is unknown
4. issue of divergence - picking wrong data associations can have catastrophic consequences

• Kalman Filter
• Particle Filter
• Graph based representations (Focus on this) - Read a tutorial on Graph-based SLAM

• Introduction
• Bayes Filter
• Occupancy Grid Maps
• Motion Model
• Observation Models
• EKF Localization
• Monte Carlo Localization
• Iterative Closest Point
• Graph-based SLAM
• Visual Features and RANSAC

I took this course due to an upcoming project in my company. I attended every Lecture and finished each Assignment. I believe this is one of the best online courses on mobile robotics. Prof. Stachniss explains so well.

I have provided my assignments here, which I completed entirely independently during the last two weeks without any external help. I enjoyed the course, assignments in particular and as of May 2021, I know of no online solutions for MSR-2 assignments except mine. My answers are not efficient ones but certainly correct. If you clone/download it, find an efficient solution, don't forget to raise an issue or a pull request.

Thanks for passing by!