robotic sensor networks - university of minnesota · •ugv can mule uav between deployment...

1
Roman Ripp [email protected] Joshua Vander Hook [email protected] Pratap Tokekar [email protected] Volkan Isler (PI) [email protected] Robotic Sensor Networks http://rsn.cs.umn.edu/ We thank Daniel Kaiser for his help with the experiments. Work is supported by NSF Awards #0917676, #1317788, #1111638. Patrick Plonski [email protected] UAV has limited battery life. Problem: Find a tour for the UAV to visit the maximum number of input points, subject to a battery life constraint. UGV can mule UAV between deployment locations. There is a trade-off between energy spent visiting points and energy spent landing/taking-off. We show how to model this capability of the UGV to mule the UAV in the form of a metric graph G. Existing orienteering algorithm when applied to G guarantees that at least ¼ of optimal number of points will be visited. Resulting UAV tour may consist of multiple ascents/descents. UAV+UGV UAV-only Car-like steering with DC motor drive Flat, known surface Max speed is a function of turning radius 1. Given a path, find minimum energy velocity profile 5m 35m 70m 100m P. Tokekar, N. Karnad, V. Isler. “Energy-Optimal Trajectory Planning for Car-Like Robots.” Autonomous Robots, 2014. acceleration velocity time 2. Given a start and destination, find energy optimal path and velocity profile Closed form velocity profiles, given a path. Build graph with discretized [x,y,θ,v] interpolate [x,y,θ] with circular arcs interpolate [v] with closed-form profiles. P. Tokekar, J. Vander Hook, D. Mulla and V. Isler. “Sensor Planning for a Symbiotic UAV and UGV system for Precision Agriculture.” IROS. 2013. David Mulla [email protected] Obtaining a ground measurement may take non-zero time. Problem: Find a tour for a given set of disks, visiting at least one point in each while minimizing travel + measurement time. Traveling Salesperson Problem with Neighborhoods (TSPN) minimizes only travel time. We present an O(r max /r min ) approximation algorithm. Standard TSPN tours may be required to take O(n) measurements. Precision agriculture will improve crop productivity through improved management of farm inputs. Detailed and updated maps of soil and crop status can help practitioners apply right amounts of fertilizer at right times and places. Data collection is a crucial component. Manually gathering data can be tedious and time-consuming. Satellite and remote sensing is costly and infrequent. Low Cost Robotic System with On-demand Sensing: UAV takes aerial images, UGV takes ground measurements. Prior measurements are combined into nitrogen map using Gaussian Process regression. Points are labeled based on the estimated nitrogen levels. For each point, we compute probability of being mislabeled. The goal is to obtain additional aerial & ground measurements at points where probability of being mislabeled is more than desired. For each point we compute a disk within which a measurement must be obtained. High, medium, low nitrogen labels Probability of being Mislabeled Obtain new measurements here Prior Nitrogen map Constructed Heightmap for Alumni Center (meters) P. A. Plonski, P. Tokekar, V. Isler. Energy-efficient Path Planning for Solar Powered Mobile Robots.” Journal of Field Robotics, 2013. P. A. Plonski, J. Vander Hook, V. Isler. “Environment and Solar Map Construction for Solar-Powered Robots.” 2014. P. A. Plonski, V. Isler. “A Competitive Online Algorithm for Exploring a Solar Map.” ICRA, 2014. McNamara Alumni Center Photovoltaic solar panels allow field robots to operate for longer. Solar maps allow solar robots to spend more time in the sun. Gaussian Process regression provides good short-term accuracy (left). Over longer time scales, we estimate the geometry of shadow-casting objects in the environment (right) using solar panel measurements. We present a solar exploration strategy with O(log n) competitive ratio in distance, where n is the number of shadow-casting objects.

Upload: others

Post on 25-Jun-2020

6 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Robotic Sensor Networks - University of Minnesota · •UGV can mule UAV between deployment locations. •There is a trade-off between energy spent visiting points and energy spent

Roman [email protected]

Joshua Vander [email protected]

Pratap [email protected]

Volkan Isler (PI)[email protected]

Robotic Sensor Networkshttp://rsn.cs.umn.edu/

We thank Daniel Kaiser for his help with the experiments. Work is supported by NSF Awards #0917676, #1317788, #1111638.

Patrick [email protected]

• UAV has limited battery life.• Problem: Find a tour for the UAV to visit the maximum number of input

points, subject to a battery life constraint.• UGV can mule UAV between deployment locations.• There is a trade-off between energy spent visiting points and energy spent

landing/taking-off.• We show how to model this capability of the UGV to mule the UAV in the

form of a metric graph G.• Existing orienteering algorithm when applied to G guarantees that at least ¼

of optimal number of points will be visited.

Resulting UAV tour may consist of multiple ascents/descents.

UAV+UGV UAV-only

Car-like steering with DC motor driveFlat, known surface

Max speed is a function of turning radius

1. Given a path, find minimum energy velocity profile

5m

35m

70m

100m

P. Tokekar, N. Karnad, V. Isler. “Energy-Optimal Trajectory Planning for Car-Like Robots.” Autonomous Robots, 2014.

acceleration velocity time

2. Given a start and destination, find energy optimal path and velocity

profile

• Closed form velocity profiles, given a path.• Build graph with discretized [x,y,θ,v]• interpolate [x,y,θ] with circular arcs• interpolate [v] with closed-form profiles.

P. Tokekar, J. Vander Hook, D. Mulla and V. Isler. “Sensor Planning for a Symbiotic UAV and UGV system for Precision Agriculture.” IROS. 2013.

David [email protected]

• UAV has limited battery life.• Problem: Find a tour for the UAV to visit the maximum number of input

points, subject to a battery life constraint.• UGV can mule UAV between deployment locations.• There is a trade-off between energy spent visiting points and energy spent

landing/taking-off.• We show how to model this capability of the UGV to mule the UAV in the

form of a metric graph G.• Existing orienteering algorithm when applied to G guarantees that at least ¼

of optimal number of points will be visited.

Resulting UAV tour may consist of multiple ascents/descents.

UAV+UGV UAV-only

P. Tokekar, J. Vander Hook, D. Mulla and V. Isler. “Sensor Planning for a Symbiotic UAV and UGV system for Precision Agriculture.” IROS. 2013.

• Obtaining a ground measurement may take non-zero time.• Problem: Find a tour for a given set of disks, visiting at least one point in

each while minimizing travel + measurement time.• Traveling Salesperson Problem with Neighborhoods (TSPN) minimizes only

travel time.• We present an O(rmax/rmin) approximation algorithm.

Standard TSPN tours may be required to take O(n) measurements.

• Precision agriculture will improve crop productivity through improved management of farm inputs.

• Detailed and updated maps of soil and crop status can help practitioners apply right amounts of fertilizer at right times and places.

• Data collection is a crucial component.• Manually gathering data can be tedious and time-consuming.• Satellite and remote sensing is costly and infrequent.

Low Cost Robotic System with On-demand Sensing: UAV takes aerial images, UGV takes ground measurements.

• Prior measurements are combined into nitrogen map using Gaussian Process regression.

• Points are labeled based on the estimated nitrogen levels.

• For each point, we compute probability of being mislabeled.

• The goal is to obtain additional aerial & ground measurements at points where probability of being mislabeled is more than desired.

• For each point we compute a disk within which a measurement must be obtained.

High, medium, low nitrogen labels

Probability of being Mislabeled

Obtain new measurements here

Prior Nitrogen map

Constructed Heightmap for Alumni Center (meters)

P. A. Plonski, P. Tokekar, V. Isler. “Energy-efficient Path Planning for Solar Powered Mobile Robots.” Journal of Field Robotics, 2013.P. A. Plonski, J. Vander Hook, V. Isler. “Environment and Solar Map Construction for Solar-Powered Robots.” 2014.P. A. Plonski, V. Isler. “A Competitive Online Algorithm for Exploring a Solar Map.” ICRA, 2014.

McNamara Alumni Center

• Photovoltaic solar panels allow field robots to operate for longer.• Solar maps allow solar robots to spend more time in the sun.• Gaussian Process regression provides good short-term accuracy (left).• Over longer time scales, we estimate the geometry of shadow-casting

objects in the environment (right) using solar panel measurements.• We present a solar exploration strategy with O(log n) competitive ratio

in distance, where n is the number of shadow-casting objects.