[SOLVED] ANP Project 1

General

The purpose of this project is to get the student(s) closely familiar with state of the art approaches on one of the topics covered in (or related to) the course.

Note: The project is an excellent opportunity to investigate a potential topic/direction for research thesis. Those interested are encouraged to get in touch with the lecturer before choosing the project topic.

Guidelines and Requirements

1. Choose a topic from the table below.
   Doodle poll: Specify your preference [link to Google Drive Doc] (first-come first- served basis, please do not alter previously-made selections)

   Deadline: April 27th 2022, no extensions, better to decide earlier

2. Tasks (more details below)
   1. (a)  Choose a topic and read paper(s) from the chosen topic.
   2. (b)  (Partial) implementation as well as method’s extension/generalization are encouraged and required to get full credit. We also encourage relevant online demonstrations using real sensors and robots.
   3. (c)  Oral presentation of the topic during the last two weeks of the semester: 20 minutes,

      unless otherwise mentioned.

   4. (d)  Submit a report that summarizes the paper(s) presented in class.

      Deadline: Submission is due July 21th 2022 (original Moed A final exam).

Written Report:

The report should summarize the material, as you comprehend it. The summary should highlight what you consider to be the main contribution of the paper(s), but should not be a “copy-paste”

from the original paper(s). The basic structure of the report should typically be as follows:

1. Introduction and overview: Introduce the paper(s) topic. Describe how the paper fits in with the contents of this course, provide a brief background (literature review) and explain why the problem is important.
2. Preliminary material and problem formulation: Present a description of relevant nota- tions and definitions, define mathematically the problem addressed by the paper(s), and summarize any preliminary mathematical material used in the paper(s).
3. Main contribution: A detailed discussion of the main results of the paper(s). This should include both a qualitative discussion and a mathematical presentation (i.e. show proofs, preferably in your own style).
4. Implementation: Demonstrate the main results of the paper(s) using simulation and/or real-world experiments. You are free to choose the programming language as well as using open source software. This also includes testing the approaches under different conditions than those originally assumed in the paper(s), as well as extending approaches to unsup- ported settings/scenarios.

1

5. Discussion and Conclusions: Summarize the report and provide some criticism: identify weak points, unrealistic assumptions or aspects that could be improved and suggest possible directions (or extensions) for future research.

The report should not exceed 10-15 pages in length. The usage of LaTex is highly recom-

mended for writing the report. Note: to get full credit, items 4 and 5 should be addressed extensively.

Oral Presentation:

The oral presentation is complementary to the written report. The presentation should be in a “lecture” style format; i.e., you will present this in front of the class with the goal of “teaching” the main points of the paper(s). It is therefore should be well organized such that participants can easily understand the key concepts. Unless otherwise mentioned, the presentation should be around 20 minutes long, with additional time allocated for questions (up to 5 minutes). The

general format of the talk should mirror the structure of the written report. All team members should take active part in the presentation, e.g. for teams of two members, each member should talk about 10 minutes.

Topics & Papers

References

Table 1: Reading material sorted by topic.

1. [1]  A.-A. Agha-Mohammadi, S. Chakravorty, and N. M. Amato. FIRM: Sampling-based feed- back motion planning under motion uncertainty and imperfect measurements. Intl. J. of Robotics Research, 33(2):268–304, 2014.
2. [2]  A. Bry and N. Roy. Rapidly-exploring random belief trees for motion planning under un- certainty. In IEEE Intl. Conf. on Robotics and Automation (ICRA), pages 723–730, 2011.
3. [3]  Panpan Cai, Yuanfu Luo, Aseem Saxena, David Hsu, and Wee Sun Lee. Lets-drive: Driving in a crowd by learning from tree search. In Robotics: Science and Systems (RSS), 2019.

2

4. [4]  Khen Elimelech and Vadim Indelman. Simplified decision making in the belief space using belief sparsification. Intl. J. of Robotics Research, 2021. Accepted, ArXiv: https://arxiv.org/abs/1909.00885.
5. [5]  E. Farhi and V. Indelman. ix-bsp: Incremental belief space planning. arXiv preprint https://arxiv.org/pdf/2102.09539, 2021.
6. [6]  Johannes Fischer and Omer Sahin Tas. Information particle filter tree: An online algorithm for pomdps with belief-based rewards on continuous domains. In Intl. Conf. on Machine Learning (ICML), Vienna, Austria, 2020.
7. [7]  G. A. Hollinger and G. S. Sukhatme. Sampling-based robotic information gathering algo- rithms. Intl. J. of Robotics Research, pages 1271–1287, 2014.
8. [8]  V. Indelman, L. Carlone, and F. Dellaert. Planning in the continuous domain: a generalized belief space approach for autonomous navigation in unknown environments. Intl. J. of Robotics Research, 34(7):849–882, 2015.
9. [9]  Peter Karkus, David Hsu, and Wee Sun Lee. Qmdp-net: Deep learning for planning under partial observability. In Advances in Neural Information Processing Systems (NIPS), pages 4694–4704, 2017.
10. [10]  D. Kopitkov and V. Indelman. No belief propagation required: Belief space planning in high-dimensional state spaces via factor graphs, matrix determinant lemma and re-use of calculation. Intl. J. of Robotics Research, 36(10):1088–1130, August 2017.
11. [11]  Mustafa Mukadam, Jing Dong, Xinyan Yan, Frank Dellaert, and Byron Boots. Continuous- time gaussian process motion planning via probabilistic inference. Intl. J. of Robotics Re- search, 37(11):1319–1340, 2018.
12. [12]  S. Pathak, A. Thomas, and V. Indelman. A unified framework for data association aware robust belief space planning and perception. Intl. J. of Robotics Research, 32(2-3):287–315, 2018.
13. [13]  Timothy Patten, Wolfram Martens, and Robert Fitch. Monte carlo planning for active object classification. Autonomous Robots, 42(2):391–421, 2018.
14. [14]  S. Prentice and N. Roy. The belief roadmap: Efficient planning in belief space by factoring the covariance. Intl. J. of Robotics Research, 28(11-12):1448–1465, 2009.
15. [15]  David Silver and Joel Veness. Monte-carlo planning in large pomdps. In Advances in Neural Information Processing Systems (NIPS), pages 2164–2172, 2010.
16. [16]  T. Smith and R. Simmons. Heuristic search value iteration for pomdps. In Conf. on Uncer- tainty in Artificial Intelligence (UAI), pages 520–527, 2004.
17. [17]  Ke Sun, Brent Schlotfeldt, George J Pappas, and Vijay Kumar. Stochastic motion planning under partial observability for mobile robots with continuous range measurements. IEEE Trans. Robotics, 2020.
18. [18]  Zachary Sunberg and Mykel Kochenderfer. Online algorithms for pomdps with continuous state, action, and observation spaces. In Proceedings of the International Conference on Automated Planning and Scheduling, volume 28, 2018.
19. [19]  WT Teacy, Simon J Julier, Renzo De Nardi, Alex Rogers, and Nicholas R Jennings. Obser- vation modelling for vision-based target search by unmanned aerial vehicles. In Intl. Conf. on Autonomous Agents and Multiagent Systems (AAMAS), pages 1607–1614, 2015.
20. [20]  Javier Velez, Garrett Hemann, Albert S Huang, Ingmar Posner, and Nicholas Roy. Modelling observation correlations for active exploration and robust object detection. J. of Artificial Intelligence Research, 2012.
21. [21]  Nan Ye, Adhiraj Somani, David Hsu, and Wee Sun Lee. Despot: Online pomdp planning with regularization. JAIR, 58:231–266, 2017.

3