4_serge_ITS for drones.pdf
0 likes76 views
The document discusses a three-level approach to drone traffic management, focusing on self-organization strategies inspired by pedestrian and vehicular traffic models. It emphasizes the importance of decentralized control and cooperation among drones to optimize interactions and improve efficiency. The research agenda aims to explore future directions and applications of these concepts in managing increasing drone traffic demands.
![Drone swarm control simulation example
▪ Simple control strategy based on
constant speed and direction strategy
inspired by our NOMAD pedestrian model
▪ No central control: drones behave based
on position information of other drones
(via sensing, communication)
▪ Example: two clouds of drones crossing
▪ Model shows safe and efficient operations:
▪ Minimum distance between drones stays
above set distance value 𝑑!
▪ Efficient passing of drones (lowest mean
speed 83% of undisturbed speeds)
▪ Test scenarios show self-organisation (like
pedestrian flows); requires more research
following of this section, we will derive a relation for the equilibrium velocit
the density gradient r⇢, that will be given by expressions (15) and (16) respe
3.2. Model derivation
We will start with the anisotropic social force model of Helbing [1] that d
influenced by the pedestrians j near pedestrian i:
E
ai =
E
v0
i E
vi
⌧i
Ai
X
j
e
Rij
Bi · E
nij ·
✓
i + (1 i)
1 + cos ij
2
◆
where Rij denotes the distance between pedestrians i and j; E
nij is the unit vect
the angle between the direction of i and the position of j; E
vi denotes the velo
The vector E
v0
i denotes the desired velocity of pedestrian i. It describes both
walking direction. We will assume that the desired velocity is a function of t
a priori. That is, we assume that there is no direct influence of the prevailing
it is, for example, only based on the shortest route to the destination. For app
refer to Refs. [21,30].
The parameter ⌧i describes the relaxation time, reflecting the time needed t
the interaction strength; Bi denotes a scaling parameter, describing the rate
function of distance; 0 i 1 denotes the anisotropy parameter, reflecti
front of i compared to those at the back. Note that i = 1 implies isotropy, i
from the back as to stimuli from the front.
As mentioned, we will assume that the system is in equilibrium, that is, w
velocity satisfies:
E
vi = E
v0
i ⌧iAi
X
j
e
Rij
Bi · E
nij ·
✓
i + (1 i)
1 + cos ij
2
◆
.](https://image.slidesharecdn.com/4sergeitsfordrones-221011162052-578bf860/85/4_serge_ITS-for-drones-pdf-11-320.jpg)
4_serge_ITS for drones.pdf
- 2. Taking surface traffic management to the next dimension… ITS for Drone Traffic
- 3. ▪ Serge Hoogendoorn Traffic & Crowd Management Victor Knoop Traffic Flow Theory Sascha Hoogendoorn-Lanser Mobility Innovation Centre Delft
- 4. Contributions of talk We aim to inspire by: ▪ Present our three level approach to distributed connected traffic management ▪ Explore with you – the drone experts – the opportunities this provides for drone management ▪ Present a research agenda and future directions In our vision, we look at a future where the airspace is becoming scarce due to the number of drones that want to use it… We look at algorithms, not so much at technology… We build on over 25 years of (research and practical) experience in traffic modelling, simulation, and management in various domains… 09-10-2020
- 5. Application examples by our group ▪ Use of simple control strategy Pedestrian flow modelling and crowd management Modelling cyclist and pedestrians in shared space Control schemes for connected and autonomous vessels Control schemes for CAVs Generic machinery: Differential game theory and dedicated numerical solution algorithms and distributed control
- 6. ▪ Our distributed control approaches are inspired on phenomena that we see in vehicular, pedestrian and vessel traffic…
- 7. Self-organisation in traffic ▪ Consider situation where two pedestrian groups walk in the opposite direction of each other ▪ So-called bi-directional lanes will form automatically without any divine intervention or central coordination, but also with limited communication between the pedestrians ▪ This process is super efficient: ▪ It occurs almost instantaneously ▪ Walking speeds remain high ▪ Very limited loss of walkway capacity (max. pedestrian / hour that can be served) ▪ This efficient self-organisation is not limited to bi-directional or crossing pedestrian flows…
- 9. Wow, modelling these interactions must be super complex! ▪ Well… No, they are not! ▪ Assuming that pedestrians… ▪ Want to get from A to B ▪ Do not want to get too close to each other and try stick to constant speed & direction ▪ Predict behaviour of others (e.g., they stay on course, do no instantaneously stop; they cooperate or not) and using differential game theory results in mathematical models describing the behaviour of individual pedestrians in relation to others ▪ Resulting simulation models reproduce efficient self-organisation accurately! 09-10-2020
- 10. ▪ Our proposition: Our game theoretical approach can inspire control approaches for efficient interactions between drones capatilising on self-organised phenomena
- 11. Drone swarm control simulation example ▪ Simple control strategy based on constant speed and direction strategy inspired by our NOMAD pedestrian model ▪ No central control: drones behave based on position information of other drones (via sensing, communication) ▪ Example: two clouds of drones crossing ▪ Model shows safe and efficient operations: ▪ Minimum distance between drones stays above set distance value 𝑑! ▪ Efficient passing of drones (lowest mean speed 83% of undisturbed speeds) ▪ Test scenarios show self-organisation (like pedestrian flows); requires more research following of this section, we will derive a relation for the equilibrium velocit the density gradient r⇢, that will be given by expressions (15) and (16) respe 3.2. Model derivation We will start with the anisotropic social force model of Helbing [1] that d influenced by the pedestrians j near pedestrian i: E ai = E v0 i E vi ⌧i Ai X j e Rij Bi · E nij · ✓ i + (1 i) 1 + cos ij 2 ◆ where Rij denotes the distance between pedestrians i and j; E nij is the unit vect the angle between the direction of i and the position of j; E vi denotes the velo The vector E v0 i denotes the desired velocity of pedestrian i. It describes both walking direction. We will assume that the desired velocity is a function of t a priori. That is, we assume that there is no direct influence of the prevailing it is, for example, only based on the shortest route to the destination. For app refer to Refs. [21,30]. The parameter ⌧i describes the relaxation time, reflecting the time needed t the interaction strength; Bi denotes a scaling parameter, describing the rate function of distance; 0 i 1 denotes the anisotropy parameter, reflecti front of i compared to those at the back. Note that i = 1 implies isotropy, i from the back as to stimuli from the front. As mentioned, we will assume that the system is in equilibrium, that is, w velocity satisfies: E vi = E v0 i ⌧iAi X j e Rij Bi · E nij · ✓ i + (1 i) 1 + cos ij 2 ◆ .
- 12. Drone interaction strategies in relation to Sensing and Communication quality Cooperative schemes Drones jointly optimise interactions and objectives Non-cooperative schemes Each drone optimises objectives without cooperation Risk-averse schemes Drone assumes other drones try to come as close as possilbe Nash game Princess monster game Coalition games Increased levels of cooperation lead to increased efficiency at the expense of higher cost of data collection and communication between drones Decreasing sensing and communication requirements
- 13. ▪ If fully decentralised self-organisation is so efficient…. (why) do we need to intervene?
- 14. Limits to self-organisation: some well know examples Freezing by heating Heterogeneity messes up efficiency Undesireable capacity distribution Priority for big flows Breakdown & capacity drop Self-organisation fails in oversaturated conditions
- 15. Limits to self-organisation make sensible interventions necessary 09-10-2020
- 16. Example interventions in connected traffic management Freezing by heating Heterogeneity messes up efficiency Undesireable capacity distribution Priority for major flow Breakdown & capacity drop Self-organisation fails in oversaturated conditions Homogenisation, e.g. by speed limits Redistribution of capacity, e.g. using traffic lights Area inflow regulation, e.g. via gating, re-routing, or demand management
- 17. Connected traffic control ▪ Example shows how in a setting with connected vehicles, an intersection controller deals with the interactions at a potential pinch point including priorisation ▪ Approach deals with joint optimisation of vehicle trajectories (for comfort, fuel consumption, and / or emission optimisation) and capacity distribution (for fair and optimal throughput) ▪ Approach substantially reduces delay (and pollution) compared to simple controller (-20%) ▪ Concept is generic and can be used for any type of traffic flow, including drones Liu M., Zhao J., Hoogendoorn S., Wang M. A single-layer approach for joint optimization of traffic signals and cooperative vehicle trajectories at isolated intersections (2022) Transp. Research Part C: Emerging Technologies, 134
- 18. Three level approach to (drone) traffic management Safe and efficient drone interaction strategies via game theoretical concepts Methods for control and prioritise interactions of flows at bottlenecks Network control methods for fully utilising network capacity becomes necessary Increasing traffic demand
- 19. Need for network management ▪ Local control cannot ensure good use of network capacity automatically ▪ From surface traffic we know that (crowds, car networks) that local congestion can lead to severe underperformance of the network ▪ Integrated Network Management (INM) aims to better distribute traffic over network (e.g., by signal coordination, smart routing) and ensure that network is not oversaturated (gating) ▪ Various approaches (decentralised, distributed, or centralised) succesfully developed and deployed in a number of scientific and applied projects (e.g., Praktijkproef Amsterdam, POC DVM Utrecht, AFM Rotterdam) To keep traffic flows at ‘t Goylaan smooth given priorities for bike and bus, a network approach was piloted in Utrecht Catastrophy at the Loveparade (2011) started with a local bottleneck which was not managed correctly
- 20. Flow and self-organisation characteristics, requirements for interventions Identify technical possibilities & requirements (lsensing, communication), regulations, algorithm development, simulation Technical possibilities & requirements (coop. sensing, communication), algorithm development (multi obj. optimisation), simulation Identification of technical requirements (communication quality, cloud solutions), security, development of coordinated control schemes Network-wide 3D traffic flow characteristics, need for network control Approach seems to have strong linkages to 5G/6G architecture! Micro interaction control Local cooperative control Network control On-device Edge Cloud
- 21. Summary Based on our experience in dynamic traffic, crowd and maritime management, we: ▪ Presented three level approach to connected drone traffic management (drone level, bottleneck level, network level) ▪ Discussed approaches and results at all levels ▪ Presented a research agenda and future directions ▪ Provided first results of taking our work ’to the next dimension’ Major opportunities expected by cross- fertilisation between our fields! 09-10-2020 Reducing urban congestion via drone delivery Using drones for automated traffic surveillance