Assignment #4
Provide a short discussion of each of the assigned papers (listed under Course Materials). Below are some questions to think about:
RRT-Connect
Optional section: 4 (but try to understand what is being proved)
- Why don't we just use A* search?
- How is the simple one-directional RRT different (and preferable to) a "random walk" in the configuration space?
- What happens in this algorithm if a collision-free path does not exist?
- There is no explicit "heuristic" in this search, why does it work well (it does)?
- The authors mention "path smoothing", what is that and why is it important in this context?
Factored-Transition Systems
- Make sure that you watch the videos. Look for subgoaling, that is, doing something to enable some other goal, as opposed to being "reactive".
- Can we just frame this as an RRT search in a bigger configuration space that involves the poses on the robot and the objects?
- Why are conditional samplers needed?
- This approach can be implemented on a real robot (video) for some tasks. What is being assumed about the task? Think about the environment classification from Russell & Norvig.
Contact-Invariant Optimization
- Think about the connection between Section 1.1 in this paper and the ideas in the "Factored Transition" paper.
- What are the advantages/disadvantages of sampling-based methods vs optimization-based methods for planning motions?
- Compare and contrast Sections 4.3, 4.4 in this paper to STRIPS planning and Sims's evolutionary approach.
- How would this approach scale to doing manipulation in the presence of many objects, as in the "Factored Transition" paper?
- No one has managed to get this approach to work on a real robot; what do you think might make that hard?
Upload a single PDF file through Stellar by
Feb 21 at 10 am.