ENEE 769M: Robotics Project
Optimal Motion Planning for Autonomous UAVs under Spatial and Finite Timing Constraints

 Usman A. Fiaz   ( UID: 115515284,   Email: [email protected] )

Updates and Correspondence (as of 03/21/2019):

Please see the replies accordingly: 


1. Will you show all the 7 bullets listed on the WWW page?
Yes, I plan on doing all 7 bullets. I'll at least do one solution approach from the three, I've listed there. It may seem a bit tall for a commitment, but I have some part of it done already so, it is do-able in this semester I guess. 

2. Is your approach centralized or decentralized? 
It is decentralized. For this semester project, I may consider some limited-information based consensus i.e., the location of an agent known to another etc. (which can be sensed in real-time) as well, to simplify the problem.

3. Are you considering a large number of UAVs? 
Eventually yes. But, for this project, I'll just use two or three UAVs.

4. What are the intrinsic benefits of MTL?
MTL lets you specify tasks in a mathematically abstract way, which can explicitly express finite timing constraints. For example, think of a task where: the robot needs to rescue a person from location A to B in 10 seconds while avoiding an obstacle along the way. This specification can be written down as a MTL formula, and then we have some theories and approaches to guarantee that this task can be accomplished, which come from formal methods for CPS. These problems can be then attacked via timed-automata, or MILP, or C-space approaches.

5. Regarding the paper that Prof. Manocha shared (ArXiv):
Yes, you sent that paper earlier. Yes, I intend to do a comparison. If possible, maybe we can combine the two approaches as well. I'll have to do some more study about this though.





Notice: This​ page is unlisted on the actual website, and is only accessible to visitors via direct sharing. Use of this page and the information that is posted, or will be posted here in future, is only for the course instructor's knowledge, and for correspondence. 
Optimal Motion Planning for Autonomous UAVs under Spatial and Finite Timing Constraints

 Usman A. Fiaz   ( UID: 115515284,   Email: [email protected] )

Goal: A formal architecture or framework for optimal (in the sense of control theory) motion planning for a team of multi-rotor UAVs, with spatial (w.r.t. space, i.e., collision avoidance for instance), and finite-timing constraints. The task will be represented as Metric Temporal Logic (MTL) specification.  The key idea of this approach is to combine space and time constraints, to develop a robust mission plan for UAVs, by treating them as hybrid Cyber-Physical Systems. This way, we can use formal methods to analyze and guarantee certain properties and specific behavior of the robots as well.

Motivation and Related Work: This work is part of my ongoing PhD research. A lot of work exists in literature, regarding motion planning for UAVs, as well as in terms of developing formal methods for guaranteeing certain behavior of simple dynamical systems. However, the novelty of this work lies in using full hybrid dynamics of the multirotor UAVs, and using MTL instead of LTL to specify complex missions. This allows the inclusion of finite time constraints on task completion, which is the case in real life and time critical scenarios. Another intention is also to work closely with Prof. Manocha; as his group has a different approach for motion planning and obstacle avoidance for a team of UAVs. I believe, I can combine, compare, and may as well perform a multi-criterion optimization to choose the better approach among the two, under different circumstances. 

Tentative Plan: ​Here is a brief list of things, I plan to do (subject to change as we go along):
1. Define a sample or benchmark mission with a team of UAVs; for instance a search and rescue scenario.
2. Define the specific task, which will be represented as a MTL formula.
3. Define the workspace: i.e., admissible areas, obstacles, boundary, and starting and finishing points.
4. Formulate the problem as an optimal control problem with suitable cost, hybrid system dynamics, and subject to task specs.
5. Define/explore solution strategies; i.e., MILP, Timed-Automata Approach, C-Space methods etc.
6. Perform a simple simulation for the chosen approach.
7. Comparison among two or more approaches.

* Will try to modify 4-7 to incorporate/compare the approach used/suggested by Prof. Manocha. Also, I am more of an application oriented person, so I plan to implement and experiment with these methods in the ARC Lab, that I am currently developing. It may not happen this semester, but I am aiming for that in the Summer, and in the Fall.




Notice: This​ page is unlisted on the actual website, and is only accessible to visitors via direct sharing. Use of this page and the information that is posted, or will be posted here in future, is only for the course instructor's knowledge, and for correspondence.