The Autonomous Land Vehicle (ALV)
`Preliminary Road-Following
`Demonstration
`
`Lowrie, James, Thomas, Mark, Gremban, Keith, Turk,
`Matthew
`
`James W. Lowrie, Mark Thomas, Keith Gremban, Matthew Turk, "The
`Autonomous Land Vehicle (ALV) Preliminary Road-Following Demonstration,"
`Proc. SPIE 0579, Intelligent Robots and Computer Vision IV, (11 December
`1985); doi: 10.1117/12.950819
`Event: 1985 Cambridge Symposium, 1985, Cambridge, United States
`
`Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 15 Feb 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
`
`PROCEEDINGS OF SPIE
`
`SPIEDigitalLibrary.org/conference-proceedings-of-spie
`
`VWGoA EX1039
`U.S. Patent No. 11,208,029
`
`

`

`Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 15 Feb 2021
`Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
`
`The Autonomous land vehicle (ALV) preliminary road-following demonstration
`The Autonomous land vehicle (ALV) preliminary road -following demonstration
`
`James W. Lowrie, Mark Thomas, Keith Gremban, Matthew Turk
`James W. Lowrie, Mark Thomas, Keith Gremban, Matthew Turk
`
`Martin Marietta Denver Aerospace Advanced Automation Technology Section
`Martin Marietta Denver Aerospace Advanced Automation Technology Section
`PO Box 179 Denver CO 80201
`PO Box 179 Denver CO 80201
`
`The autonomous land vehicle program overview
`The autonomous land vehicle program overview
`
`The ALV project is sponsored by the Defense Advanced Research Project Agency (DARPA) as
`The ALV project is sponsored by the Defense Advanced Research Project Agency (DARPA) as
`part of its Strategic Computing Program and contracted through the Army Engineer Topo(cid:173)
`part of its Strategic Computing Program and contracted through the Army Engineer Topo-
`graphic Laboratories (ETL) under contract DACA76-84-C-0005. The purpose of the strategic
`The purpose of the strategic
`graphic Laboratories (ETL) under contract DACA76 -84 -C -0005.
`computing program is to advance the state of the art in artificial intelligence, image
`computing program is to advance the state of the art in artificial intelligence, image
`understanding, and advanced computer architectures and to demonstrate the applicability of
`understanding, and advanced computer architectures and to demonstrate the applicability of
`these technologies to advanced military systems.1
`these technologies to advanced military systems.l
`
`The strategic computing (SC) program is separated into three primary areas technology
`The strategic computing (SC) program is separated into three primary areas -- technology
`base, applications, and infrastructure. The technology base contractors are tasked with
`base, applications, and infrastructure.
`The technology base contractors are tasked with
`pursuing generic long-range high-payoff research in numerous disciplines including image
`pursuing generic long -range high -payoff research in numerous disciplines including image
`understanding, expert systems, planning and reasoning, symbolic processing architectures,
`understanding, expert systems, planning and reasoning, symbolic processing architectures,
`high-speed signal processing systems, and others. The application areas are being funded
`high -speed signal processing systems, and others.
`The application areas are being funded
`to transition the technology from the research domain to the military application domain
`to transition the technology from the research domain to the military application domain
`with the intent of demonstrating a series of progressively more complex operational capa(cid:173)
`with the intent of demonstrating a series of progressively more complex operational capa-
`bilities. Finally, the infrastructure of the SC program provides the framework for both
`Finally, the infrastructure of the SC program provides the framework for both
`bilities.
`the research community and the application programs. This framework includes information
`the research community and the application programs.
`This framework includes information
`networks, research machines, and system development tools.
`networks, research machines, and system development tools.
`
`The ALV project is one of the SC program's application areas aimed at advancing and
`The ALV project is one of the SC program's application areas aimed at advancing and
`demonstrating the state of the art in autonomous navigation and tactical decisionmaking.
`demonstrating the state of the art in autonomous navigation and tactical decisionmaking.
`The project is driven by the series of progressively more difficult demonstrations identi(cid:173)
`The project is driven by the series of progressively more difficult demonstrations identi-
`fied in Table 1. These successive demonstrations were selected because they drive the
`These successive demonstrations were selected because they drive the
`fied in Table 1.
`development of technology in artificial intelligence, image understanding, and advanced
`development of technology in artificial intelligence, image understanding, and advanced
`computer architectures.
`computer architectures.
`Table 1. ALV demonstration.
`Table 1. ALV demonstration.
`
`Year
`Year
`May 1985
`May 1985
`(Preliminary
`(Preliminary
`Road-Following
`Road -Following
`Demonstration)
`Demonstration)
`November 1985
`November 1985
`(Road-Following
`(Road -Following
`Demonstration)
`Demonstration)
`
`1986
`1986
`(Obstacle
`(Obstacle
`Avoidance
`Avoidance
`Demonstration)
`Demonstration)
`1987
`1987
`(Crosscountry
`( Crosscountry
`Demonstration)
`Demonstration)
`
`Distance
`Distance
`1 km
`1 km
`
`5km
`5 km
`
`20 km
`20km
`
`Speed
`Speed
`5km/h
`5 km /h
`
`Capability
`Capability
`The vehicle will traverse a uniform road with smooth curves at a constant speed.
`The vehicle will traverse a uniform road with smooth curves at a constant speed.
`During conditions where the vision subsystem is unable to locate the road, the
`During conditions where the vision subsystem is unable to locate the road, the
`vehicle may follow a prestored map of the track. The vehicle must navigate from
`vehicle may follow a prestored map of the track. The vehicle must navigate from
`visual data over 75% of the distance.
`visual data over 75% of the distance.
`10 km/h
`The vehicle will traverse a nonuniform road with sharp corners. The vehicle
`10 km /h The vehicle will traverse a nonuniform road with sharp corners. The vehicle
`speed will vary as a function of vision confidence and road geometry. The vehicle
`speed will vary as a function of vision confidence and road geometry. The vehicle
`must navigate from visual data 100% of the time. The vehicle must demonstrate an
`must navigate from visual data 100% of the time. The vehicle must demonstrate an
`autonomous counter-rotate capability.
`autonomous counter -rotate capability.
`20 km/h
`20 km /h The vehicle will traverse a nonuniform road with numerous intersections. The
`The vehicle will traverse a nonuniform road with numerous intersections. The
`vehicle must sense and model obstacles placed on the road surface and plan a path
`vehicle must sense and model obstacles placed on the road surface and plan a path
`to avoid them.
`to avoid them.
`
`10 km
`10km
`
`5 km/h
`5 km /h
`
`The vehicle must be capable of planning an a priori route through the terrain using
`The vehicle must be capable of planning an a priori route through the terrain using
`a prestored terrain database. The system must then use sensory data to model the
`a prestored terrain database. The system must then use sensory data to model the
`local terrain and avoid natural obstacles placed along the route. The position of
`local terrain and avoid natural obstacles placed along the route. The position of
`the vehicle with respect to the route must be monitored and updated as necessary.
`the vehicle with respect to the route must be monitored and updated as necessary.
`The vehicle must navigate through rough roadways.
`The vehicle must navigate through rough roadways.
`
`Success of the ALV project depends on careful coordination with the technology base
`Success of the ALV project depends on careful coordination with the technology base
`contractors to transfer technology from the research domain to the application domain as
`contractors to transfer technology from the research domain to the application domain as
`rapidly as possible. To simplify the technology transition process, ALV was designed as a
`To simplify the technology transition process, ALV was designed as a
`rapidly as possible.
`The intent
`flexible testbed that will enable rapid transition from hypothesis to testing. The intent
`flexible testbed that will enable rapid transition from hypothesis to testing.
`of the testbed is to encourage the technology base contractors to conduct experiments with
`of the testbed is to encourage the technology base contractors to conduct experiments with
`the vehicle in a realistic environment. The results of these experiments would then lead
`The results of these experiments would then lead
`the vehicle in a realistic environment.
`naturally into design of the demonstration system.
`naturally into design of the demonstration system.
`
`This paper describes the long-range ALV system concept that the project is building
`This paper describes the long -range ALV system concept that the project is building
`toward, the system requirements for road-following, gives an overview of the ALV system as
`toward, the system requirements for road -following, gives an overview of the ALV system as
`it was configured for the May 1985 demonstration, and contains detailed descriptions of
`it was configured for the May 1985 demonstration, and contains detailed descriptions of
`the vision, navigator, pilot, electronic, and vehicle subsystems.
`the vision, navigator, pilot, electronic, and vehicle subsystems.
`
`336 / SPIE Vol. 579 Intelligent Robots and Computer Vision (1985)
`336 / SPIE Vol 579 Intelligent Robots and Computer Vision (1985)
`
`

`

`Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 15 Feb 2021
`Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
`
`ALV long-range system concept
`ALV long -range system concept
`
`The progressive system demonstration schedule, along with the requirement to transition
`The progressive system demonstration schedule, along with the requirement to transition
`capabilities from the strategic computing technology base contractors, makes it essential
`capabilities from the strategic computing technology base contractors, makes it essential
`to define a long-range generic system architecture. It is more beneficial to build each
`It is more beneficial to build each
`to define a long -range generic system architecture.
`demonstration system within the framework of the long-range system architecture than to
`demonstration system within the framework of the long -range system architecture than to
`discard each demonstration system following its completion. By defining a long-range
`By defining a long -range
`discard each demonstration system following its completion.
`system architecture and analyzing the long-term requirements, we can project the technol(cid:173)
`system architecture and analyzing the long -term requirements, we can project the technol-
`ogy voids that will become the topic for research by the technology base contractors.
`ogy voids that will become the topic for research by the technology base contractors.
`
`Definition of the long-range system architecture has been the topic of a series of
`Definition of the long -range system architecture has been the topic of a series of
`working group meetings between various technology base contractors and the ALV project
`working group meetings between various technology base contractors and the ALV project
`team. 2 The following technology base contractors participated in this definition
`The following technology base contractors participated in this definition- -
`team.2
`University of Maryland, Carnegie Mellon University, SRI International, Advanced Informa(cid:173)
`University of Maryland, Carnegie Mellon University, SRI International, Advanced Informa-
`tion and Decision Systems, Hughes AI Center, and Honeywell.
`tion and Decision Systems, Hughes AI Center, and Honeywell.
`
`Autonomous mobility in a dynamic unstructured environment requires that a system sense
`Autonomous mobility in a dynamic unstructured environment requires that a system sense
`its environment, model critical features using the sensed data, reason about the model to
`its environment, model critical features using the sensed data, reason about the model to
`determine a mobility path, and control the vehicle along that path. Evaluation of these
`determine a mobility path, and control the vehicle along that path.
`Evaluation of these
`basic mobility requirements resulted in the definition of the system concept shown in
`basic mobility requirements resulted in the definition of the system concept shown in
`Figure 1. Two additional requirements associated with the objective of the strategic
`Figure 1.
`Two additional requirements associated with the objective of the strategic
`computing program were factored into this configuration. First, the primary emphasis of
`computing program were factored into this configuration.
`First, the primary emphasis of
`the program is on perception and reasoning with minimal research being pursued in the
`the program is on perception and reasoning with minimal research being pursued in the
`areas of control and physical vehicles. Because it is also desirable to rapidly integrate
`areas of control and physical vehicles.
`Because it is also desirable to rapidly integrate
`and test numerous concepts on the testbed vehicle, we have defined a "virtual vehicle"
`and test numerous concepts on the testbed vehicle, we have defined a "virtual vehicle"
`consisting of the physical vehicle, the sensors, and the control subsystems. The hardware
`consisting of the physical vehicle, the sensors, and the control subsystems.
`The hardware
`and software interfaces at this level are well known and experiments that conform to these
`and software interfaces at this level are well known and experiments that conform to these
`interfaces can be rapidly integrated and tested.
`interfaces can be rapidly integrated and tested.
`
`Knowledge Base
`Knowledge Base
`- Digital Terrain Database
`- Digital Terrain Database
`- Long-Term Scene Model
`- Long -Term Scene Model
`- Vehicle State Estimate
`- Vehicle State Estimate
`
`Queries A Priori
`Queries A Priori
`Model
`Model
`
`Scene
`Scene
`Model
`Model
`
`Queries
`Queries
`
`Vehicle
`Vehicle
`State
`State
`
`Model
`Model
`Parameters
`Parameters
`
`Acquisition
`Acquisition
`Commands
`Commands
`
`Sensors
`Sensors
`
`Data
`Data
`
`Perception
`Perception
`
`Task
`Task
`Request
`Request
`
`Task
`Task
`Status
`Status
`
`Reflexive
`Reflexive
`Scene
`Scene
`Model
`Model
`
`Reference
`Reference
`Trajectory
`Trajectory
`
`Vehicle State
`Vehicle State
`Update
`Update
`
`Control
`Control
`
`Reasoning
`Reasoning
`
`Vehicle
`Vehicle
`State
`State
`Mission
`Mission
`Plan&
`Plan &
`Status
`Status
`
`Mission
`Mission
`Goals
`Goals
`
`Human
`Human
`Interface
`Interface
`
`Figure 1. Long-range ALV system architecture.
`Figure 1. Long -range ALV system architecture.
`The human operator will specify the mission goals and constraints that should be
`The human operator will specify the mission goals and constraints that should be
`factored into decisionmaking through the man/machine interface (MMI). The complexity of
`The complexity of
`factored into decisionmaking through the man /machine interface (MMI).
`these mission goals will increase with each successive demonstration. For May 1983 the
`For May 1983 the
`these mission goals will increase with each successive demonstration.
`goal specification was simply to follow the road for 1 km. In 1987 the goal becomes more
`In 1987 the goal becomes more
`goal specification was simply to follow the road for 1 km.
`complex travel to point A, perform task B, proceed to point C, .
`.
`Later demonstra-
`.
`Later demonstra(cid:173)
`complex -- travel to point A, perform task B, proceed to point C,
`tions will also include complex tactical situations that must be dealt with.
`tions will also include complex tactical situations that must be dealt with.
`
`.
`
`The reasoning system will interpret these mission goals and decompose them into the
`The reasoning system will interpret these mission goals and decompose them into the
`operations to be performed by the vision subsystem. As part of this decomposition
`As part of this decomposition
`operations to be performed by the vision subsystem.
`process, the reasoning subsystem will access a digital terrain database being developed by
`process, the reasoning subsystem will access a digital terrain database being developed by
`the Engineer Topographic Laboratories and plan an a priori route through the environment
`the Engineer Topographic Laboratories and plan an a priori route through the environment
`to achieve the mission goals. The perception system is considered to be a resource of the
`The perception system is considered to be a resource of the
`to achieve the mission goals.
`reasoning subsystem. In this capacity the reasoning subsystem will specify goals for the
`In this capacity the reasoning subsystem will specify goals for the
`reasoning subsystem.
`perception system to perform. These goals will include specification of the features of
`perception system to perform.
`These goals will include specification of the features of
`interest, a time allocation for the process, and a focus of attention defining the geome(cid:173)
`interest, a time allocation for the process, and a focus of attention defining the geome-
`tric area to be modeled. The perception subsystem will then decompose this goal into
`tric area to be modeled.
`The perception subsystem will then decompose this goal into
`
`SPIE Vol. 579 Intelligent Robots and Computer Vision (1985) / 337
`SPIE Vol 579 Intelligent Robots and Computer Vision (1985) / 337
`
`

`

`Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 15 Feb 2021
`Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
`
`specific perception tasks. The perception subsystem will have sole control over all
`specific perception tasks.
`The perception subsystem will have sole control over all
`sensors and will produce only a high-level symbolic model of the environment for
`sensors and will produce only a high -level symbolic model of the environment for
`reasoning. Figure 2 illustrates the sensor/perception interface. Following completion of
`reasoning.
`Figure 2 illustrates the sensor /perception interface.
`Following completion of
`model generation, the perception subsystem will pass the model to the reasoning subsystem
`model generation, the perception subsystem will pass the model to the reasoning subsystem
`along with a status description. The status will indicate if the perception system was
`along with a status description.
`The status will indicate if the perception system was
`able to achieve the goal and if not will describe the potential reasons for failure.
`able to achieve the goal and if not will describe the potential reasons for failure.
`Vision
`Sensors
`Vision
`Sensors
`RGB Data
`RGB Data
`(512x480)
`(512x480)
`Two high-quality color TV cameras are provided.
`Two high -quality color TV cameras are provided.
`Format size is 512x480. Vision issues an acquisi(cid:173)
`ACQ Command
`Format size is 512x480. Vision issues an acquisi-
`tion command via a software interface.
`tion command via a software interface.
`
`ACQ Command _
`
`Color TV
`
`TV cameras are mounted on independent computer-
`TV cameras are mounted on independent computer -
`controlled pan/tilt drives. Vision issues pan- and
`controlled pan /tilt drives. Vision issues pan- and
`tilt-angle commands via a software interface and can
`tilt -angle commands via a software interface and can
`read actual position.
`read actual position.
`
`Two TVs mounted on a single sensor rail can be in(cid:173)
`Two TVs mounted on a single sensor rail can be in-
`dependently positioned. Vision issues position com(cid:173)
`dependently positioned. Vision issues position com-
`mands for each sensor via a software interface and
`mands for each sensor via a software interface and
`can read actual position.
`can read actual position.
`
`Status
`Status
`(Theta Actual)
`(Theta Actual)
`Theta Commands
`Theta Commands
`
`Status
`Status
`P1,P2 Commands
`P1, P2 Commands
`
`Status
`Status
`Theta Commands
`Theta Commands
`
`RGB Data
`RGB Data
`
`Pan /Tilt
`1
`
`Interocular
`Control
`
`Pan /Tilt
`2
`
`Color TV
`2
`
`Range
`Range
`(256x128)
`(256x128)
`0.5, 0.65f 0.85,1.5,10.0 M
`0.5, 0.65, 0.85, 1.5, 10.Oµ
`ACQ Command
`ACQ Command
`
`Multispectral
`Scanner
`
`Multispectral scanner provides range data and radio-
`Multispectral scanner provides range data and radio-
`metric data in the
`. 5-/i, 0.65-M, 0.85-ju, 1,5-/z and
`metric data in the .5 -p, 0.65 -p, 0.85 -p, 1.5 -p and
`10.0-A/ bands (preliminary). Vision issues an acquisi(cid:173)
`10.0 -µ bands (preliminary). Vision issues an acquisi-
`tion command via a software interface.
`tion command via a software interface.
`
`Other
`Sensors
`
`(Other sensors are processed outside the
`(Other sensors are processed outside the
`vision subsystem.)
`vision subsystem.)
`Figure 2 Sensor/Perception Interface
`Figure 2 Senso /Perception Interface
`For the 1987 time frame we anticipate there will be two forward-looking high-resolution
`For the 1987 time frame we anticipate there will be two forward -looking high -resolution
`color TVs mounted on independent pan/tilt mounts with a controllable interocular distance
`color TVs mounted on independent pan /tilt mounts with a controllable interocular distance
`ranging from 1 to 5 ft. Each camera will have an independent 3-channel 8-bit digitizer.
`ranging from 1 to 5 ft.
`Each camera will have an independent 3- channel 8 -bit digitizer.
`A 5-channel multispectral laser scanner being developed by the Environmental Research
`A 5- channel multispectral laser scanner being developed by the Environmental Research
`Institute of Michigan (ERIM) will also be incorporated.
`Institute of Michigan (ERIM) will also be incorporated.
`
`The reasoning subsystem will interpret the perception model and will plan a path far the
`The reasoning subsystem will interpret the perception model and will plan a path for the
`vehicle to avoid nontraversable regions and localized obstacles. Because of the signifi(cid:173)
`Because of the signifi-
`vehicle to avoid nontraversable regions and localized obstacles.
`cant amount of time involved in processing the sensory data to produce a symbolic model,
`cant amount of time involved in processing the sensory data to produce a symbolic model,
`it is not possible in the near future for the vehicle control system to close the high(cid:173)
`it is not possible in the near future for the vehicle control system to close the high-
`speed servoloop from visual data. Therefore we have introduced the concept of a reference
`Therefore we have introduced the concept of a reference
`speed servoloop from visual data.
`trajectory whereby the vehicle control system follows a selected path from one model until
`trajectory whereby the vehicle control system follows a selected path from one model until
`the next model is generated. Figure 3 illustrates the reasoning control interface portion
`Figure 3 illustrates the reasoning control interface portion
`the next model is generated.
`of the virtual vehicle. The control subsystem will be responsible for three activities.
`The control subsystem will be responsible for three activities.
`of the virtual vehicle.
`First, it will control the motion of the vehicle along the specified trajectory. Second,
`Second,
`First, it will control the motion of the vehicle along the specified trajectory.
`the control subsystem will evaluate the specified trajectory and determine such unsafe
`the control subsystem will evaluate the specified trajectory and determine such unsafe
`conditions as sudden high-speed turns. Third, the vehicle state estimate consisting of
`Third, the vehicle state estimate consisting of
`conditions as sudden high -speed turns.
`vehicle position, velocity, heading, pitch, and roll will be maintained within the control
`vehicle position, velocity, heading, pitch, and roll will be maintained within the control
`subsystem.
`subsystem.
`
`The physical vehicle consists of a drive chassis supplied by Standard Manufacturing, a
`The physical vehicle consists of a drive chassis supplied by Standard Manufacturing, a
`122-hour auxiliary power unit, and a 60,000-Btu air conditioner. This physical platform
`122 -hour auxiliary power unit, and a 60,000 -Btu air conditioner.
`This physical platform
`is considered to be sufficient to support the electronics for all projected demonstrations
`is considered to be sufficient to support the electronics for all projected demonstrations
`and experiments. Figure 4 illustrates the physical vehicle.
`Figure 4 illustrates the physical vehicle.
`and experiments.
`
`System requirements for the May 1985 demonstration
`System requirements for the May 1985 demonstration
`
`The May 1985 demonstration required the vehicle to autonomously travel on a paved road
`The May 1985 demonstration required the vehicle to autonomously travel on a paved road
`over a distance of 1 km at a speed of 5 km/hour. The 1-km distance requirement introduced
`over a distance of 1 km at a speed of 5 km /hour.
`The 1 -km distance requirement introduced
`the need for a robust vision subsystem capable of operating on hundreds of successive
`the need for a robust vision subsystem capable of operating on hundreds of successive
`scenes. The 5-km/hour speed requirement introduced the need for special-purpose computers
`scenes.
`The 5 -km /hour speed requirement introduced the need for special -purpose computers
`that could rapidly process imagery. This section summarizes the analyses conducted to
`This section summarizes the analyses conducted to
`that could rapidly process imagery.
`define the system requirements for the May 1985 demonstration.
`define the system requirements for the May 1985 demonstration.
`
`338 / SP/E Vol. 579 Intelligent Robots and Computer Vision (1985)
`338 / SPIE Vol 579 Intelligent Robots and Computer Vision (1985)
`
`

`

`Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 15 Feb 2021
`Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
`
`Control law behavior can be modified
`Control law behavior can be modified
`by adjusting control parameters.
`by adjusting control parameters.
`
`Planning generates a time-tagged
`Planning generates a time -tagged
`sequence of reference points in the
`sequence of reference points in the
`vehicle coordinate system at the time
`vehicle coordinate system at the time
`of the last LIMS update.
`of the last LNS update.
`
`LIMS data are available at a 40-ms
`LNS data are available at a 40 -ms
`interval.
`interval.
`
`Planner is responsible for issuing up
`Planner is responsible for issuing up-
`dates to the LNS to meet system per(cid:173)
`dates to the LNS to meet system per-
`formance req u i rements.
`formance requirements.
`
`Figure 3. Reasoning control interface.
`Figure 3. Reasoning control interface.
`
`Control Parameter Modifiers
`Control Parameter Modifiers
`
`Reference
`Trajectory
`
`Position &
`Head ng
`Error
`
`i
`Lat, Lon, Az, El, Roll
`
`}
`
`LNS Updates
`
`Control
`Law
`
`Land
`Navigation
`System
`
`Reasonableness
`Checking
`
`VLl
`
`1t
`
`VR
`
`ALV
`
`0000
`
`Vehicle
`Vehicle
`Health &
`Health &
`Status
`Status
`Interface
`Interface
`
`I
`I
`
`.
`
`,5gitflfpflSf|S:3% /
`. ^,/fs^m^iiS^Mff^.
`
`Vehicle
`Vehicle
`Control
`Control
`nterface
`Interface
`
`Rack 1
`Rack 1
`Navigation
`Navigation
`Planner
`Planner
`
`Air-
`Air -
`Conditioning-
`Conditioning
`Ducts
`Ducts
`
`Rack 2
`Rack 2
`Vision
`Vision
`
`Laser
`Laser
`Scanner
`Scanner
`
`Evaporator/
`Evaporator/
`Cooler
`Cooler
`
`Cable Trays
`Cable Trays
`
`Rack 3
`Communications
`Communications
`Equipment
`Equipment
`
`Environmental..
`Environmental
`Control Unit
`Control Unit
`F igu re 4. Physical veh icle chassis.
`Figure 4. Physical vehicle chassis.
`Access Panel
`Access Panel
`The test track is 6 -m wide on the average and the vehicle is 3 -m wide.
`The test track is 6-m wide on the average and the vehicle is 3-m wide. To remain on the
`To remain on the
`road, the vehicle must maintain a lateral error no greater than +1.5 m.
`road, the vehicle must maintain a lateral error no greater than +1.5 m. To provide a
`To provide a
`it was decided that the vehicle must travel within +0.5 m of the road
`margin of safety it was decided that the vehicle must travel within +0.5 m of the road
`margin of safety
`centerline as shown in Figure 5.
`centerline as shown in Figure 5.
`
`As an additional safety factor it was decided that the vehicle could navigate off a
`As an additional safety factor it was decided that the vehicle could navigate off a
`prestored map of the road whenever the vision subsystem produced low-confidence models as
`prestored map of the road whenever the vision subsystem produced low- confidence models as
`long as visual information was used more than 75% of the time. To implement this vision
`long as visual information was used more than 75% of the time.
`To implement this vision
`override capability the entire test track was surveyed to an accuracy of 0.15 m. When
`override capability the entire test track was surveyed to an accuracy of 0.15 m.
`When
`
`SPIE Vol. 579 Intelligent Mobots and Computer Vision (1985} / 339
`SP /E Vol. 579 Intelligent Robots and Computer Vision (1985) / 339
`
`

`

`Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 15 Feb 2021
`Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
`
`- Speed
`- Speed
`- Distance
`- Distance
`- Accuracy
`- Accuracy
`- Competency
`- Competency
`
`5 km/h
`5 km /h
`1 km
`1 km
`±0.5 m from Centerline
`±0.5 m from Centerline
`Vision Scene Model Used for Navigation
`Vision Scene Model Used for Navigation
`More Than 75% of the Time
`More Than 75% of the Time
`
`Vision Override
`Vision Override
`- Test track centerline has been surveyed to an
`- Test track centerline has been surveyed to an
`accuracy of 0.15 m.
`accuracy of 0.15 m.
`- If the vehicle wanders outside a ±0.5-m error band,
`- If the vehicle wanders outside a ±0.5 -m error band,
`the map data will be used to bring the vehicle back
`the map data will be used to bring the vehicle back
`to the centerline.
`to the centerline.
`- Vision override is expected to be used in conditions
`- Vision override is expected to be used in conditions
`where the vision subsystem cannot locate the road
`where the vision subsystem cannot locate the road
`edges.
`edges.
`
`±0.5-m Error Band
`Error
`Band
`±0.5 -m
`___L
`f~T
`
`/
`
`/
`
`/
`
`Road
`Center line
`Centerline
`
`'
`I
`/
`/
`/
`Figures. May demonstration performance requirements.
`Figure 5. May demonstration performance requirements.
`conditions did not allow the vision subsystem to segment the road in the image or derive
`conditions did not allow the vision subsystem to segment the road in the image or derive
`the 3-D geometry of the road edges, the map data were used to control the vehicle. Vision
`Vision
`the 3 -D geometry of the road edges, the map data were used to control the vehicle.
`override is an artifact of the May 1985 demonstration only and will not be incorporated in
`override is an artifact of the May 1985 demonstration only and will not be incorporated in
`future demonstration systems.
`future demonstration systems.
`
`To control the vehicle on a continuous basis at a fixed velocity we believed that the
`To control the vehicle on a continuous basis at a fixed velocity we believed that the
`vision-based scene model could not be generated at the servoloop update rates. Analysis
`Analysis
`vision -based scene model could not be generated at the servoloop update rates.
`indicated that the servoloop needed to operate at 40-ms update intervals and that the
`indicated that the servoloop needed to operate at 40 -ms update intervals and that the
`state-of-the-art vision algorithms were three orders of magnitude slower than that rate.
`state -of- the -art vision algorithms were three orders of magnitude slower than that rate.
`Therefore the concept of a reference trajectory was used (Fig. 6). This concept allows
`This concept allows
`Therefore the concept of a reference trajectory was used (Fig. 6).
`the servoloop to operate at 40 ms while the visual information is updated at a much slower
`the servoloop to operate at 40 ms while the visual information is updated at a much slower
`rate. At time TQ the vehicle acquires the i^1 image while the control system steers
`At time To the vehicle acquires the ith image while the control system steers
`rate.
`from the (i-l) tn trajectory. The i*-*1 image is processed up to time T]_ to generate a
`The ith image is processed up to time T1 to generate a
`from the (i -1)th trajectory.
`scene model and corresponding traj