My Research
I am a post doctoral researcher
in the Department of
Mechanical Engineering at the
Current Research
I am currently
working on the following projects:
(A)
Learning
for Locomotion (with Dan Lee, Vijay Kumar, George Pappas and Dan
Koditschek)
The Learning
for Locomotion project aims to develop controllers for quadruped walking on
rough terrain by applying techniques from machine learning. We use a standard
robotic platform called Littledog developed by Boston Dynamics, Inc. to develop
our learning controllers. We decouple the problem of sensing the terrain from
the problem of control by using a motion capture system and laser scanned
terrains. This gives us complete information about the nature of the terrain.
Learning has been applied to two specific areas in our
approach: (1) offline learning to determine the right places to place the feet
based on local terrain features and (2) trial to trial learning to refine the
path chosen for the robot. The offline learning for determining where to put
the feet is carried out by estimating (from trials performed using heuristics
for choice of foothold) the amount of slip corresponding to a particular choice
of footholds and using data from multiple trials to score footholds.
The controller used for the gait follows a crawl gait
characterized by having at least three feet on the ground at all times. The
feet are phased a ¼ cycle apart. A reactive control module checks for
deviations from the prescribed/desired motion of the body of the robot and
applies corrective actions to ensure that the robot moves to its desired pose
before the next command is executed. The reactive control module has been
successful in allowing the robot to recover from some very bad poses.
The following movies show trials over different terrains:
(a) Movie
1 (60 MB)
(b) Movie 2
(75 MB)
(c) Movie
3 showing climbing motion onto a 15 cm step.
(d) Movie
4 showing the robot getting up from a fallen-down position.
Proprioceptive
localization for a quadrupedal robot on known terrain
We present a novel method for the localization of a legged
robot on known terrain using only proprioceptive sensors such as joint encoders
and an inertial measurement unit. In contrast to other proprioceptive pose
estimation techniques, this method allows for global localization (i.e.,
localization with large initial uncertainty) without the use of exteroceptive
sensors. This is made possible by establishing a measurement model based on the
feasibility of putative poses on known terrain given observed joint angles and
attitude measurements. Results are shown that demonstrate that the method
performs better than dead-reckoning, and is also able to perform global
localization from large initial uncertainty. This
movie (60 MB) explains the process while more details can also be found in
an ICRA paper (This
work was performed in collaboration with Paul Vernaza, Roman Geykhman and
Daniel Lee).
(B)
CKBot:
Highly mobile modular robots (with Mark Yim)
This
project aims to develop a modular robot that consists of many reconfigurable
modules and demonstrate its multifunctional capabilities and reconfiguration in
a desert for running, climbing, structuring, life-protecting, and flying in a
micro-gravity environment. We have built a first generation module with a
single degree of freedom and multiple connection ports on different faces.
Follow the link to the project page to see gaits for several different
configurations.
Our
first generation hardware module has a single servo motor and a dedicated PIC
onboard. Communication between modules and a master controller is through a
global CAN bus. Power is supplied through a tether. Modules can connect to each
other on four different faces in different orientations. We are presently implementing
an IR based serial communication protocol for local communication. Our next
generation modules will include sensors (accelerometers, etc.) and onboard
power supply. We are also exploring alternative module designs including leg
modules.
Automatic Configuration Recognition Methods in
Modular Robots
Recognizing modular
robot configurations composed of hundreds of homogeneous or heterogeneous
modules is a significant challenge. Matching a new modular robot configuration
to a library of known configurations helps in identifying and applying control
schemes on the new configuration and is a step towards self-repair. In this paper we
presented three different algorithms to address the problem of (a) matching and
(b) mapping a new robot configuration onto a known library of configurations.
The first method solves the problem using graph isomorphisms and can identify
configurations that share the same underlying graph structure, but have different
port connections amongst the modules. The second technique utilizes graph
spectral techniques and proves useful in finding approximate or near matches to
configurations that may not be in the library. The third algorithm exploits the
unique structure of the problem for the particular robots used in our research
to achieve impressive gains in performance and speed over existing techniques,
especially for larger configurations. Together, these three algorithms provide
a complete solution to the problem of configuration matching and mapping.
Results and examples are provided to compare the performance of the three
algorithms and discuss their relative advantages. (This work was performed in
collaboration with Michael Park, Alex Teichman and Mark Yim).
Dynamic Rolling for a Modular
Reconfigurable
modular robots have the ability to use different gaits and configurations to
perform various tasks. A rolling gait is the fastest currently implemented gait
available for traversal over level ground and shows dramatic improvements in
efficiency. In this paper,
we analyzed and implemented a sensor-based feedback controller to achieve
dynamic rolling for a loop robot. The robot senses its position relative to the
ground and changes its shape as it rolls. This shape is such that its center of
gravity is maintained to be in front of its contact point with the ground, so
in effect the robot is continuously falling and thus accelerates forward. Using
simulation and experimental results, we show how the desired shape can be
varied to achieve higher terminal velocities. The highest velocity achieved in
this work is 26 module lengths per second (1.6m/s) which is believed to be the
fastest gait yet implemented for an untethered modular robot. One of the major
findings is that more elongated shapes achieve higher terminal velocities than
rounder shapes. We demonstrate that this trend holds going up as well as down
inclines. We show that rounder shapes have lower specific resistance and are
thus more energy efficient. The control scheme is scalable to an arbitrary
number of modules, shown here using 8 to 14 modules. (This work was performed
in collaboration with Jimmy Sastra and Mark Yim).
Previous Work: Modular
Locomotion Systems (PhD thesis)
A copy of my thesis can be found
here (PDF).
The main focus of my PhD thesis research was the control and dynamics of modular robot systems. I worked with Dr. Vijay Kumar in the area of modular locomotion systems. Part of this research was also conducted with Dr. Jim Ostrowski (now with Evolution Robotics). There is a wide variety of definitions and interpretations of the word modular. There are an equally large number of modular robots made in universities and research labs around the world. Our focus is on modular mobile robots. We consider such robots to consist of a base with a number of ports. A module is attached at each port. The ports allow flow of information and power from the base to the modules. The modules are locomotion elements including powered wheels, legs and passive wheels.
In motion planning for such a
modular robot, we take a top-down
approach to the problem. We start with a motion plan for the base or central
platform of the robot and propagate it down to the individual modules. The base
does not explicitly control every little detail of the motion of the modules.
Instead, the module only gets information about what kind of motion the robot
needs to execute. A controller on each module then figures out the motion the
module needs to perform for the whole robot to move as desired. This approach
allows us to easily pull out modules and insert new ones. More details on this
research can be found in [18] where we presented a motion planning scheme for a
hybrid robot with wheels and legs. The results presented in [18] are applicable
to kinematic locomotion systems, i.e. systems where the inputs are velocity
inputs. The next stage of this research, described below, is to study dynamic
systems, i.e. systems where kinematic analysis is insufficient to explain the
motion of the system or design control laws for the systems.
Dynamic Modular Locomotion systems - Rollerblading Robot
(Click here to go to the Rollerblading Robot page)
Previous Work: RoboTrikke
This is a novel underactuated locomotion system with a single input. The system has passive wheels and moves forward when the front steering axis is given a periodic input. By offsetting this input in a particular direction, the system can be made to undergo a net rotation as well. In our paper [2] based on this system, we showed that it is possible to steer this system to follow (on average) a straight line trajectory. Check out our cool movie on this system. Also check out the new movie for a new version of this robot with a rider on the vehicle.
Previous Work: Biking without pedaling
Having tried out the Trikke, we
realized that a bicycle is basically a 3D version of the Trikke. It must be
possible to ride a bicycle without pedaling- in fact, a lot of people can stay
stationary on their bikes for a long time by moving only a very small distance.
Based on this idea, we proceeded to model and control a bike without pedals!
Bicycles have been studied for a long time and there are well-known models of bicycle dynamics. We developed our own model and compared it to recently published benchmark results to confirm their accuracy. Using ideas from the control of an underactuated system called the Acrobot and the sinusoidal actuation method used for the RoboTrikke, we developed a controller to ride a bike without pedaling. Subsequently, Mike Park, a graduate student at GRASP, demonstrated this on a real bicycle (warning: Windows Media Version :-( ).
Previous Work: Robot Soccer
I also worked with Dr. Jim Ostrowski on the UPennalizers, a team of Sony four-legged robot dogs programmed to play Soccer. More information on my participation on this team and links to the team website can be found on my Robot Soccer Page. We wrote a paper on omnidirectional walking based on this research [19]. We also have a couple of team reports based on this research [8], [9].
Publications
Journal Papers/Book
Chapters/Technical Reports
1. Sachin Chitta and Vijay Kumar, “Biking Without Pedaling”, in preparation for the ASME Journal of Mechanical Design.
2. Sachin Chitta, Frederik Heger and Vijay Kumar, “Design, Analysis, Simulation and Experimental Results for a Rollerblading Robot”, in preparation for the International Journal of Robotics Research.
3. Sachin Chitta, Peng Cheng and Vijay Kumar, “RoboTrikke: A Novel Undulatory Locomotion System”, in preparation for the International Journal of Robotics Research.
4. Michael Park, Sachin Chitta, Alex Teichman and Mark Yim, “Automatic Configuration Recognition Methods in Modular Robots”, under review for the International Journal of Robotics Research.
5. Jimmy Sastra, Sachin Chitta and Mark Yim, “Dynamic Rolling for a Modular Loop Robot”, under review for the International Journal of Robotics Research.
6.
Sachin Chitta, “Dynamics and Control of a Class of Modular
Locomotion Systems”, Department of Mechanical Engineering and Applied
Mechanics,
7.
Sachin
Chitta and J. Ostrowski,
“Technical Report MS-CIS-01-08: Enumeration and Motion Planning for
Modular Mobile Robots”, Technical Report, Department of Computer and
Information Science,
8. Sachin Chitta, William Sacks, Jim Ostrowski, Aveek Das, and P. K. Mishra., “The University of Pennsylvania RoboCup Legged Soccer Team”, in A. Birk, S. Coradeschi, S. Tadokoro (Editors), Lecture Notes in Computer Science 2377: RoboCup 2001: Robot Soccer World Cup V, Springer Verlag, 2001. (PDF)
9. Jim P. Ostrowski, Ken A. McIsaac, Aveek Das, Sachin Chitta, and Julie Neiling, “The University of Pennsylvania RoboCup Legged Soccer Team”, in P. Stone, T. Balch, G. Kraetzschmar (Editors), Lecture Notes in Computer Science 2019: RoboCup 2000: Robot Soccer World Cup IV, Springer Verlag, 2000. (PDF)
Refereed Conference Papers
10.
Sachin Chitta, Paul Vernaza
and Daniel Lee, “Proprioceptive Localization for a Quadrupedal Robot on
Known Terrain”, accepted to the 2007 IEEE International Conference on
Robotics and Automation,
11. Sachin Chitta, Mustafa Karabas, Kevin Galloway and
Vijay Kumar, “RoboTrikke: Design, Modeling and Experimentation with a
Robotic Trikke”, in Proceedings of the 2006 ASME Design Engineering
Technical Conference,
12. Jimmy Sastra, Sachin Chitta and Mark Yim,
“Dynamic Rolling for a Modular Loop Robot”, in Proceedings of the
International Symposium on Experimental Robotics, Rio Di Janerio, Brazil, 2006.
(PDF)
13.
Sachin
Chitta, and Vijay Kumar,
"Biking Without Pedaling", in the International Design Engineering
and Technology Conference,
14.
Sachin
Chitta, Peng Cheng, Emilio
Frazzoli and Vijay Kumar, "RoboTrikke: A Novel Undulatory Locomotion
System ", in the IEEE International Conference on Robotics and Automation,
15.
Sachin
Chitta, Frederik Heger and
Vijay Kumar, “Design, Analysis, Simulation and Experimental Results for a
Rollerblading Robot”, in Proceedings of the 2004 ASME Design
Engineering Technical Conference,
16. Sachin Chitta, Frederik Heger and Vijay Kumar,
“Design and Gait Control of a Rollerblading Robot”, in Proceedings
of the 2004 IEEE International Conference on Robotics and Automation,
17.
Sachin
Chitta and Vijay Kumar,
"Dynamics and Generation of Gaits for a Planar Rollerblader", in
Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and
Systems,
18. Sachin Chitta and James P. Ostrowski, “Motion Planning for Heterogeneous Modular Mobile Systems”, in Proceedings of the 2002 IEEE International Conference on Robotics and Automation, Washington D.C., 2002. (PDF)
19. Sachin Chitta and James P. Ostrowski, “New Insights into Quasi-Static and Dynamic Omnidirectional Quadruped Walking”, in Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, Maui, Hawaii, 2001. (PDF)