Kevin Lynch's Research Interests

Distributed Estimation and Control of Multi-Agent Systems
Underwater Localization
Manipulation by Pushing
Dynamic Manipulation
Parts Feeding
Underactuated Robots
Collaborative Manipulation and Materials Handling
Haptics
Friction
Sensing for Manipulation
Ph.D. Thesis

Videos of Experiments

Decentralized Estimation and Control of Multi-Agent Systems

We are pursuing a framework for systematic design of emergent behaviors in sensing and communication networks of mobile agents. The problem is to design a control law to run on each agent, based on sensor and communication input, so that the desired collective behavior emerges. Example tasks include sensor coverage, environmental monitoring, formation control, multi-agent pursuer-evader, and other types of self-organization. The key constraints are that each agent may have significant dynamics and limited sensing, computation, motion, and communication capabilities. The behavior of the system should improve or degrade gracefully as agents are added or deleted; in other words, the approach should be scalable, robust, and require no central controller.

Our approach requires each agent to simultaneously (1) estimate properties of the global behavior of the system and (2) use those estimates in a motion control law. This suggests a systematic approach of separately designing the estimator and controller, and then ensuring that the coupled system retains desired performance properties.


Underwater Localization

We are developing novel ribbon-fin propulsors for highly-maneuverable vehicles operating in cluttered environments. To support close-range sensing in cluttered environments, we are developing an active electrosensor. This sensor creates an electric field and uses one or more detectors to detect voltage perturbations in the electric field. These perturbations are then interpreted to either find a target in the environment or to localize the vehicle in a known environment.

In the figure at right, the two squares are electrodes that create an electric field, the diamond is the voltage detector, and the circle is a perfectly conductive object in water. With Gaussian noise added to the sensor reading, a single sensor reading allows localization of the object with >68% likelihood in the regions labeled white and >95% likelihood in the white + lightest gray regions. Remaining ambiguity can be eliminated by multiple sensor readings at different locations.


Manipulation by Pushing

For planar manipulation problems, pushing provides an option for robots lacking the size, strength, or dexterity to grasp and carry an object. For example, it is usually easier to push a couch into position than to grasp and lift it. Pushing also provides a mechanism for manipulating multiple objects simultaneously. Our work on robotic pushing focuses on three aspects: References to relevant papers are given below. More details can be found on stable pushing and parts feeding using pushing.

Dynamic Manipulation

By not grasping, a simple robot with few degrees-of-freedom can control an object with more degrees-of-freedom by exploiting dynamic effects. These extra degrees-of-freedom come from manipulation phases such as controlled slipping and rolling. In contrast, a robot that carries an object with a firm grasp requires as many degrees-of-freedom as those of the object it wishes to control.

Our work on dynamic manipulation has been on motion planning, feedback control, and implementation of robotic tasks such as dynamically snatching an object from a table (using inertial forces to keep the object fixed to the robot), rolling an object on the surface of the manipulator, and throwing and catching. Nonlinear optimization is used to plan robot trajectories that achieve the desired motion via coupling forces through the nonprehensile (graspless) contact.

See our dynamic manipulation testbed Flatland. References to relevant papers are given below. See our video gallery of dynamic tasks.


Parts Feeding

Some industrial parts feeders, such as bowl feeders and the Sony APOS system, utilize vibratory dynamics to help orient parts. For instance, a vibrating pallet of holes can be used to capture parts in a desired orientation by using an appropriate shaking motion.

Our work on parts feeding has focused on feeding parts on a conveyor belt using a system that is very simple, yet still programmable to allow feeding different types of parts; designing agitation (throwing and catching or vibration) to induce desired equilibria or generalized equilibria (limit cycles) in a single part or multiple parts (e.g., throwing together an assembly); and simultaneous manipulation of one or more parts by creating controlled frictional force fields on a 6-DOF vibrating plate (see video at right and the project page here).

References to relevant papers are given below. More videos can be found here.


Underactuated Robots

An underactuated robot is a robot with fewer actuators than degrees-of-freedom. Our work on underactuated robots includes motion planning for underactuated second-order mechanical systems, controllability analysis, motion planning, and feedback control of a three joint manipulator with only two actuators, and a controllability analysis of a planar body (such as a hovercraft which glides on water with zero friction) with unilateral thrusters.

See the video of a 3 DOF robot with a passive (unactuated) joint navigating through an obstacle field.


Collaborative Manipulation and Materials Handling

Our work on assisted materials handling has focused on the idea that it is easier for a human to manipulate a load constrained to move along a (workless) frictionless guide than to manipulate the load freely. We are interested in designing guides which make the task as easy (or ergonomic) as possible for the human.

Haptics

Work in haptics includes displaying realistic dynamic virtual environments to the sense of touch, particularly systems with hard, smooth constraints. Examples include virtual linkages and other virtual devices subject to holonomic and nonholonomic constraints.

Friction

Systems with Coulomb friction may have multiple solutions (ambiguity) or no solutions (inconsistency) to their equations of motion. These properties of rigid-body mechanics with Coulomb friction are well known. However, a careful analysis of Coulomb friction in robotic manipulation reveals other unexpected results: for instance, the perfectly rough surface of classical mechanics is not equivalent to infinite Coulomb friction, a common misidentification in textbooks.

Sensing for Manipulation


Ph.D. Thesis


Back to Kevin Lynch's home page.