Title: Algorithms for Collaborating Intelligent Vehicles Krithi Ramamritham
1Algorithms for Collaborating Intelligent
Vehicles Krithi Ramamritham
- with Gurulingesh R, Vipul Shingde, Ashish Gudhe,
- Neera Sharma, Jatin Bharadia, Sachitanand
Malewar - Embedded Real-Time Systems Lab
- IIT Bombay
2Talk Outline
- Motivation
- Focus of our Work
- Automatic Merge Control (AMC) System
- Our approaches
- Optimization Formulation
- Head of Lane Approach
- Observations
- Adaptive Cruise Control (ACC)
- Our approaches
- Dual-mode system
- Two-level repository
- Observations
- Conclusions Further Work
3Motivation
- Rapid increase of computer-controlled functions
in automotive applications - e.g. New Mercedes S-class cars employ at least
70 networked ECUs - Independent black-box implementation is
practically infeasible - Cost, Integration Complexity
4Motivation
- Modern vehicles are becoming intelligent with
- Onboard sensors, Computational Communication
Resources, - ( road-side sensors)
- Increase in safety-enhancing applications
- Brake-by-Wire, Collision Avoidance system,
Adaptive Cruise Control, Intersection Merge
Control, etc - deal with critical data and deadline bound
computations - have stringent requirements on
- Freshness of data
- Completion time of tasks
5Focus of Our Work
- Efficient utilization of computing resources
- Ensuring safety comfort to the driver
- Providing real-time guarantees
- Proof of concept using robotic vehicular
platforms built in our lab
6Automatic Merge Control
- Ensure safe vehicle maneuver at 2-road
intersections - Extended to n-road
- Primary Aim
- Safety
- Secondary Aim
- Throughput ?
- Fuel Efficiency ?
- Latency ?
7AMC System
- Vehicles with
- Autonomous driving capability
- Communication ability
- Determining merge sequence
- Decentralized by individual vehicles depending
on environment data - Centralized by road-side infrastructure node
- Keeps track of Vehicle Profiles
- Determine
- Merge Sequence
- Future Behavior velocity, acceleration
- Communicate with vehicles
8Optimization Formulation
- Formulated as non-linear optimization
- Minimize average Driver Time To Intersection
(DTTI) - Minimize f
- Maximize throughput
- Minimize f
- Constraints
- Precedence Constraint
- Mutual Exclusion within Intersection
- Bound on Vehicle Velocity and Acceleration
- Safety Criteria (Safety distance S)
9Optimization Formulation
System Input ?i,j dij ,uij ,S System Output
?i,j tij
- Using tij, we determine
- Merge sequence
- Future Behavior of vehicles
- aij - Acceleration (constant) of each vehicle
- vij(t) - Velocity of each vehicle
- Acceleration commands are then sent to respective
vehicles
10Head of Lane (HoL)
- Basic Idea
- Consider head-of-lane vehicle (HoL) in each lane
- Determine which head vehicle should go first
- Consider next set of head vehicles
- Metric Average DTTI
- Accelerate vehicle whenever possible
11HoL Algorithm
- Given current profile and maximum possible
acceleration, compute when will each head vehicle
reach the merge region
No
Yes
Vehicles interfere?
Follow the above order and the future behavior
Choose winner based on preference metric
12Merge Order in case of Interference
- Nearest Head
- Allow the vehicle nearest to the merge region to
go first - Cascading Effect
- Effect of particular order on the vehicles behind
- Number of vehicles affected
- Net deceleration introduced
- Is HoL Optimal?
- HoL and Optimal have same vehicle order
13Continuous stream of vehicles
- Above algorithms applicable only for a particular
snapshot - Issues in extending
- Identifying the snapshot of vehicles
- Determining how often to capture these snapshots
(i.e., how often to run algorithm)
- Approaches
- Sporadic Approach
- Snapshot All vehicles in AoI
- Sporadicity x/Vmax where
- X - the distance of the closest vehicle outside
the AoI - Disadvantages Reconsiders all vehicles (except
those who have passed through merge region)
stability concern reconsidering the vehicles
which are nearer to merge region - Zonal Partitioning
- Assumption Initially, no vehicles in Zone1
- Apply algorithm to snapshot of vehicles in Zone2
- When vehicles from Zone3 enter Zone2, reapply the
algorithm without disturbing those vehicles which
have moved into Zone1
14Implementation
- Optimization formulation implemented in Matlab
- fmincon function
- Execution time
- Variable
- HoL approach with nearest head implemented in C
- Execution time
- Order of 0.1 second
15System Input
- System Parameters
- Amax 4m/s2 Amin -4m/s2
- Vmax 27m/s Vmin 0m/s
- S 5m
-
- Vehicle Profile at t0
16System Output
17Results Observed
Optimization Formulation
HoL Approach
18AMC -- Summary
- Safe maneuvering at intersection using our
approaches - Results of HoL are comparable to the optimal
solution (fmincons may not be global optimum) - Execution time of HoL very small compared to that
of Matlab
19AMC -- Ongoing Work
- Extend HoL n-road, cascading effect
- Extend both approaches to deal with continuous
stream of vehicles multi-zone, sporadic - Fine tuning vehicle preference, road priority
- Decentralizing both the approaches
- Challenging scenario human automated vehicles
- Develop real-time support for the system and
demonstrate the concept on robotic vehicular
platforms
20AMC -- Findings
- Both internal external factors influence the
application - Internal vehicle profile, acc/dec capabilities,
etc - External other vehicles profiles, location,
acc/dec capabilities - Merge region is a shared resource among vehicles
on different roads - Separation of responsibilities desired
- Vehicles maintain profile, make decision, etc
- Infrastructure Node provide global info, help
to make decision, etc
21Introduction to ACC
- Adaptive Cruise Control tries to maintain
- Safe Distance when there is a leading vehicle
- Set Speed when there is no leading vehicle in
its path
22Introduction to ACC
- Extension of Cruise Control.
- Operates either in
- Distance Control state
- Speed Control state
Des_Dist Host_Vel Timegap ?
where Host_Vel is Host Vehicle
velocity TimeGap is set by the driver ? for
additional safety
23Design Issues
- Effective tracking of dynamically varying data
General Practice Prepare for the Worst
Over-Sampling
24Design Issues
- Timely updates of derived data
General Practice Periodic updates
Unnecessary Updates
25Design Issues
- Some tasks will execute only in some modes
- Adapt parameters when lead car is far
- Sense adjacent lane, time to collision when car
is near
General Practice Single mode design for
simplicity
Poor CPU utilization
Scheduling Overhead
Not modular
26Our Approach
- Dual Mode System
- Two mutually exclusive phases of operation
- Safety Critical Mode
- Non Safety Critical Mode
- Current mode depends on
- Distance of Separation
- Rate of change of Distance
27Our Approach
- Real-Time Data Repository
- Two level data store
- Environment Data Repository
- Derived Data Repository
- Task Scheduling
- Constant Bandwidth Server (CBS)
- Task Scheduling
- Constant Bandwidth Server (CBS)
28Our Approach
- Real-Time Data Repository
29Robotic Vehicle Experimental Setup
- Capabilities
- Obstacle detection Range 2m
- Maximum speed 0.50 cm/s
- White-line following
30Results Observations
- Cruise Control
- Set Speed 25 cm/s
31Results Observations
2. ACC Varying Velocity - Velocity Response
32Results Observations
3. ACC Varying Velocity - TimeGap
33Results Observations
- Basic Experiments
- 1. Cruise Control
- Set Speed 25 cm/s
- 2. Adaptive Cruise Control
- Varying Velocity
- ACC tries to maintain
- Set speed when there is no leading vehicle
- Safe Distance when there is leading vehicle
- Variation in graphs due to Shaft Encoder error
34Results Observations
- Real-Time Data Repository Experiments
- Task under observation DistT (which updates DoS)
- Threshold Value 5cm
- Leading vehicle with uniform speed
Velocity Response
35Results Observations
- Real-Time Data Repository Experiments
- Task under observation DistT (which derives DoS)
- Threshold Value 5cm
- Leading vehicle with varying speed
Velocity Response
36Results Observations
- Real-Time Data Repository Experiments
- Task under observation DistT (which derives DoS)
- Threshold Value 5cm
- Time Window 0-12 sec
37Results Observations
- Real-Time Data Repository Experiments
- Task under observation DistT
- Less number of Updates
- Compared to conventional approach
- Efficient usage of computing resource
- Functionality/Safety not affected
38Results Observations
- Dual Mode Experiment
- Mode change criteria Lead Dist 65 cm/s
- Periodicity of tasks P(SC mode) ½ P(NC mode)
Velocity Response
39Results Observations
- Dual Mode Experiment
- Mode change criteria Lead Dist 65 cm/s
- Periodicity of tasks P(SC mode) ½ P(NC mode)
Timegap Response
40Results Observations
- Dual Mode Experiment
- Mode change criteria
- leading distance 65 cm/s
- Periodicity of Tasks
- P(SC Mode) ½ P(NC Mode)
- Compared to conventional approach
- Efficient usage of computing resource
- Functionality/Safety not affected
- Conservative Approach while deciding SafeDist
41ACC Conclusions
- Presented issues involved in developing real-time
support for ACC - Efficiently utilized processor capacity by
designing ACC using following concepts - Mode change
- Real-time data repository
- Provided scheduling strategies to meet timing
requirements
42ACC Ongoing Work
- More analysis of the system design (mode-change
criteria, task periodicities, threshold values,
etc.) - Application needs are being mapped to distributed
platform - Hybrid design dual-mode with data repository
- Study of controllers stability and performance
- Vehicle platooning
43AMC ACC Findings
- Both internal external factors influence the
application - Internal vehicle profile, acc/dec capabilities,
driver set timegap, etc - External other vehicles profiles, location,
acc/dec capabilities - Road is a shared resource among vehicles
- Separation of responsibilities desired
- Vehicles maintain profile, make decision, etc
- Infrastructure Node provide global info, manage
the shared resource intelligently, etc
44Ongoing Research
- CarOS
- Manager within vehicles to
- Manage processes within vehicles
- Manage resources within vehicles
- Provide networking solutions
- Provide security or protection, etc.
- Example Any OSEK standard OS with extensions
- RoadOS
- Manages the transport infrastructure
45Goals of Ongoing Research
- RoadOS
- Manager of infrastructure
- Manage resources shared by vehicles
- Provide networking solution
- Migration of vehicles, etc.
- RoadDB
- Maintains real-time information
- Tracks vehicle status and road-side
infrastructure - Real-Time Continous Query processing
- Effective data dissemination, etc.
46Goals of on-going Research
- Design and Development of
- RoadOS
- RoadDB
- CarOS (extend OSEK standard)
- Interaction between RoadOS CarOS
- Sensor Network Data Management
- Demonstrate on robotic vehicular
- platforms built in our lab.
47References
- Petros Ioannou Cheng-Chih Chien. Autonomous
Intelligent Cruise Control. IEEE Trans. On
Vehicular Technology, 42(4)657-672, 1993. - Thomas Gustafsson Jörgen Hansson. Dynamic
on-demand updating of data in real-time database
systems. In Proceedings of ACM SAC 2004. - Gerhard Fohler Flexibility in Statically
Scheduling Real-Time Systems. PhD Thesis,
Technischen Universitat Wien Austria, Apr. 1994. - L. Sha R. Rajkumar J. Lehoczky K. Ramamritham.
Mode Change Protocols for Priority-Driven
Preemptive Scheduling. In Journal of Real Time
Systems, 1(3)243-265, Dec 1989. - Tornsten Bruns Eckehard Munch. Intersection
management as self-organisation of mechatronic
systems. In Proceedings of 6th International
Heinz Nixdorf Symposium on New Trends in Parallel
and Distributed Computing, Paderborn, Germany,
2006. - Kurt Dresner and Peter Stone, Multiagent Traffic
Management An Improved Intersection Control
Mechanism, In The Fourth International Joint
Conference on Autonomous Agents and Multiagent
Systems (AAMAS 05), July 2005. - Tsugawa S. Inter-vehicle communications and
their applications to intelligent vehicles an
overview. In IEEE Intelligent Vehicle Symposium,
vol. 2, pp. 564-569, Versailles,France, 2002. - T. Uno A. Sakaguchi S. Tsugawa. A merging
control algorithm based on inter-vehicle
communication. In IEEE International Conference
on Intelligent Transportation Systems, pp.
783-787, Tokyo, Japan, 1999.
48Thank You !!
Embedded Real-Time Systems Lab Indian Institute
of Technology Bombay http//www.it.iitb.ac.in/rese
arch/labs/erts_lab/index.php