Source of seminal papers in robotics

Question

Is there a database or website that has collected the seminal papers in different disciplines of robotics like machine learning, AI, mobile robots, etc.

By seminal I mean papers that made a path-breaking impact on the theoretical side, for example, proved a theorem that captivated and inspired a large number of derivative works. So, I am mainly looking for high-impact papers that made fundamental contributions in mathematical modeling, algorithm design etc. and not so much on the hardware application side of it because in that respect videos of Boston dynamics or festo robotics are the better sources of inspiration.

I know some seminal papers like:-

1. Latombe's planning book
2. Khatib's potential field method
3. SLAM paper
4. Kalman's derivation of his filter
5. DP paper by Berketsas

But similar seminal papers are missing in say robotic formation control etc.

So my question has someone collected papers that rigorously and mathematically showed some big result in robotics.

Answer 1

Nathan Ratliff has documented some very nice papers in Control Theory and Motion Optimization. You can access them here and here, resp.

Particularly related to decision making problems, you might want to check Geoff Hollinger and Gaurav Sukhatme's course. The have a good list of readings in a logical flow. In general robotics, you may want to look at Pieter Abbeel's readings list for CS287.

Asides for that, one of the first papers my advisor recommended me was that Alberto Rodriguez on Effector Form Design. It is a beautiful paper with very appealing and seemingly useful mathematical formulation. To put his work into perspective, I think his thesis is worth a read (50 pages or so).

I think generally, papers which get like Best Paper Awards in top robotics conferences have sound theoretical grounds. Also, most graduate level courses incorporate good reading material and often these courses are openly accessible.

Answer 2

I am not aware of any online repository that particularly collects seminal papers in robotics. But I think anyone working or having worked in robotics for some time would more or less have their own collection of the so-called seminal papers in their field.

So, here are motion planning-related papers that I think are within the scope of this question. (Please feel free to edit this answer to add more content or correct what I wrote should anything is wrong.)

• Lozano-Perez, Tomas. "Spatial planning: A configuration space approach." IEEE transactions on computers 2 (1983): 108-120.

  This is kind of a paradigm shifting paper in motion planning. Previously a robot was always thought of as a 2D or 3D geometric object in the physical space. While motion planning algorithms at the time worked for problems with a 2D/3D polygonal robot moving in a 2D/3D maze and problems with an easy planar manipulator, it was very difficult to generalize such algorithms to solve complex problems consisting of a multi-DOF manipulator, for example. This paper basically says, "Hey, let's treat a robot as a point to a space (the so-called configuration space)." Then finding a path for a robot from configuration A to configuration B will simply be a problem of finding a path connecting two points in the configuration space. And now pretty much any modern planning algorithms operate in configuration space.

• Kavraki, Lydia E., Petr Svestka, J-C. Latombe, and Mark H. Overmars. "Probabilistic roadmaps for path planning in high-dimensional configuration spaces." IEEE transactions on Robotics and Automation 12, no. 4 (1996): 566-580.

  Kuffner, James J., and Steven M. LaValle. "RRT-connect: An efficient approach to single-query path planning." In Robotics and Automation, 2000. Proceedings. ICRA'00. IEEE International Conference on, vol. 2, pp. 995-1001. IEEE, 2000.

  These two papers presented ground-breaking motion planning algorithms called Probabilistic Roadmap (PRM) and Rapidly-exploring Random Tree (RRT), respectively. I would not go into details here. But basically they are so fundamental, yet practical. There are literally hundreds of variants being proposed in the literature after these two algorithms.

• Karaman, Sertac, and Emilio Frazzoli. "Sampling-based algorithms for optimal motion planning." The international journal of robotics research 30, no. 7 (2011): 846-894.

  This paper presented the very first result of (asymptotically) optimal motion planning algorithms. Novelty of this paper lies not only in the presented algorithms but also the mathematical analysis that was used to prove various properties, including optimality, of planners.

And since the question also mentioned Latombe's planning book, I would feel so wrong not mentioning the following one

• LaValle, Steven M. Planning algorithms. Cambridge university press, 2006.

Answer 3

Let us first introduce what “theorem prooving” in mathematics mean. The famous book “The art of computing programming” is devoted to that principle. The idea is, that the book has a theoretical part, for example about sorting, and after each chapter there are questions for the practical part. The students can answer the question, and if they are right, they get the mark I from the professor. Theorem proofing means, to teach only material which is already known. Sometimes the algorithm, which were presented by Knuth were around 100 years old.

What does that mean for the area of robotics? Google Scholar has indeed some papers which can be used for student evaluation. They have mostly a title starting with “formal methods” or “Proof”. The idea is, to present only knowledge which is correct and asks if the students have understand it. This suppresses innovation and prevents that the students are exploring failed robotics projects. The problem in AI is, that in most cases the knowledge is not given. That means, there is no standard way in pathplanning, so it is difficult to force the students to answer such kinds of questions. But there are many subfields in Artificial Intelligence who are well suited for proofing something, for example the AIXI theory of Marcus hutter, or the students can proof that the perceptron is able to master the XOR problem.

A quite old paper which goes into the direction is The Use of Forth in the Instruction of Introductory Robotics which is using a teaching programming language to proof on a formal level how “Learning from demonstration” works.

Answer 4

I would add Probabilistic Robotics as generally a great book to start mostly mobile robotics.

For Robotics in general, I would start with the Modern Robotics book. http://hades.mech.northwestern.edu/index.php/Modern_Robotics

Both of these summarize the state of the art at their publication date very well.