Noel duToit
Research

My Other Stuff:
Main Page
Research Page
Publications
My Photo Album
Life before Caltech
Links

Current research
My broad field of interest in mechanical engineering is controls. Controls is a rather diverse field, so there is need to be more specific. Though most of my training to date has been in classical control, my interest lie with topics such as adaptive control. This can be regarded as learning in the control theoretic sense. This has lead me to become interested in planning, from the control theoretic point of view. There is much interest in high-level, long term autonomy/ planning. This has been viewed mostly from an AI (artificial intelligence) perspective, and I am hoping to combine these approaches (AI and control theoretic planning). At least that is the idea for now. More information to follow in a couple months when I have to submit a research proposal.

In preparation for this project/topic, I am working on the Team Caltech entry into the DARPA Grand Challenge. There are roughly 50 people involved in the effort, which is being lead by Richard Murray. I am coordinating the navigation team as part of TA'ing CS/EE/ME 75, which means that I get to deligate ~15 students, but it also means that I am effectively responsible for all the planning modules, and the lower-level control systems.

Some details on the planning problem that we are facing. What makes the planning problem for urban driving challenging are the following: first, the vehicle has to generally operate in a structured environment while obeying traffic rules. This means that the vehicle has to exhibit very specific behavior in specific situations. This behavior needs to be enstilled in the vehicle. Second, the environment that the vehicle has to operate in is dynamic, including some other autonomous vehicles (the behavior of which could potentially be very unpredictable). Third, we have no a priori map of the environment, so we have to sense/construct the map. But there is always uncertainty associated with this sensed data from which the maps are constructed. There will be uncertainty about what is sensed, where it is sensed, how big it is, and what it is doing. And somehow we have to plan through this environment. Not a simple problem to solve.

Roughly, the way we are approaching the problem is to divide it up into 3 parts: route planning, corridor planning, and path planning. The route planning is kind of at the level of directions from Google Maps. This problem has mostly been solved. The corridor planning is the layer that I am working on. This is where the route goals are combined with the map information and traffic rules, and a corridor is defined that prescribes a set of 'actions'. These actions are determined by the desired behavior for the current situation. Without going into too many details, the corridor planner is divided into 3 parts: situation estimator (am I driving on a road segment, or at an intersection), the action estimator (based on my situation, what is the best/ safest action to take), and the corridor specification (based on the action and map and traffic rules, where am I allowed to drive). This corridor gets passed to a path planner that comes up with some (sub)optimal trajectory for the vehicle to follow.

I am sure that most people did not even make it to this line, but if you are interested shoot me a mail and I will be happy to give more information.


Previous research
Before starting at Caltech, I worked on a MEMS-scale piezoelectric vibration energy harvester. The main idea is that there is enough vibrational energy in the ambient to power a wireless sensor node. The power requirements for these nodes have been (and are continuing to) decline in recent years due to advances in low-power DSP's and VLSI-system design. Unfortunately, battery technology does not follow Hooke's law, as is evident in many appliances today (e.g., laptops, cellphones, etc.). Thus, an alternative, infinite power source is highly desireable. Though many forms. of ambient energy harvesting is out there, we focussed on vibration energy harvesting.

I was the first student on the project in my lab, so my project was rather high-level. The project included (among other things): investigate ambient vibration sources, develop a generic vibration energy harvester model, expand the model to the electromechanically coupled case. Next, we extended the model to the cantilevered beam structure. I also set up an experimental testbed and tested a macroscale device for model validation. I also tried my hand at the microfabrication of a micro-scopic device, but unfortunately I ran out of time. If this is of interest to you, check out the publications page.