0 likes | 14 Views
Exploring a graph using a swarm of oblivious robots poses challenges that are tackled in this work. The study focuses on determining the minimum number of robots required to effectively explore graphs of varying sizes. The concepts of distinguishable configurations and automorphisms are pivotal in understanding the lower bound for robot exploration. Theoretical frameworks and proofs are provided to establish the optimal algorithm for graph exploration tasks, specifically in the context of robotic systems with limited communication capabilities.
E N D
A Lower Bound for Graph Exploration by a Swarm of Oblivious Robots Stéphane Devismes Joint work with: Anissa Lamani, Franck Petit, and Sébastien Tixeuil
Roadmap • Models and Exploration problems • The lower bound • Application: lower bound for the ring case • Proof of the result • Optimal algorithm for the ring case GRASTA 2018, Berlin 2
Roadmap • Models and Exploration problems • The lower bound • Application: lower bound for the ring case • Proof of the result • Optimal algorithm for the ring case GRASTA 2018, Berlin 3
Robots • A swarm of K robots – Motion actuators – Visibility sensors – Uniform & anonymous – Oblivious – No communication mean GRASTA 2018, Berlin 4
Discrete environment • A simple graph G = (V,E) – Connected – Unoriented – Anonymous Possible moves Locations GRASTA 2018, Berlin 5
Discrete environment • A simple graph G = (V,E) – Connected – Unoriented – Anonymous Possible moves Locations GRASTA 2018, Berlin 6
View • Instantaneous snapshot of the system – Full or Distance d – Multiplicity (information about each location) • Weak (0, 1, or Tower) • Strong (0, 1, or Tower of x robots) GRASTA 2018, Berlin 7
View • Instantaneous snapshot of the system – Full or Distance d – Multiplicity (information about each location) • Weak (0, 1, or Tower) • Strong (0, 1, or Tower of x robots) GRASTA 2018, Berlin 8
Symmetry ? GRASTA 2018, Berlin 9
Symmetry • Worst case decision: chosen by an adversary ? GRASTA 2018, Berlin 10
Cycle • Look: Instantaneous Snapshot • Compute: Based on this observation, decides (deterministically or probabilistically) to either stay idle or move to one of the neighboring nodes • Move: Move toward the destination GRASTA 2018, Berlin 11
Models (from the stronger to the weaker) • FSYNC: each robot executes a full cycle at each step • SSYNC: at each step, a nonempty subset of robots executes a full cycle (+ fairness) • ASYNC: Look, Compute and Move are atomic however the time between Look, Compute, and Move is finite but unbounded Remark: in any snapshot, no robot on edges GRASTA 2018, Berlin 12
(Terminating) Exploration • Deterministic version: starting from any towerless configuration (i.e., at most one robot per node) – Exploration: Each node must be visited by at least one robot – Termination: Within finite time, every robot stays idle • Probabilistic version (Las Vegas): termination with probability one GRASTA 2018, Berlin 13
Challenge • What is the minimal number of robots K necessary to explore a graph of size n > K? (under a distributed scheduler) GRASTA 2018, Berlin 14
Challenge • What is the minimal number of robots K necessary to explore a graph of size n > K? (under a distributed scheduler) Termination Detection Need to use configurations as an implicit memory Oblivious Robots GRASTA 2018, Berlin 15
Roadmap • Models and Exploration problems • The lower bound • Application: lower bound for the ring case • Proof of the result • Optimal algorithm for the ring case GRASTA 2018, Berlin 16
Our result Theorem: deterministic) exploration with – K robots – on a graph of n>K nodes is possible only if • K>2 and there exists a set S of at least n-K+1 configurations such that: – any two different distinguishable, and – in every configuration in S, there is a tower of less than K robots. Terminating (probabilistic or configurations in S are GRASTA 2018, Berlin 17
Distinguishable Configurations • Two configurations C1and C2are indistinguisable if there exists an automorphism f on G = (V,E) such that ∀p∈V, Multiplicity(C1,p) = Multiplicity(C2,f(p)) • If C1and C2are not indistinguishable, then they are distinguisable • Automorphism: f: G → G such that ∀p,q∈V, {p,q}∈E ⟺ {f(p), f(q)}∈E GRASTA 2018, Berlin 18
Example of indistinguishable configurations a b c f(i) f(h) f(g) d e f f(f) f(e) f(d) g h i f(c) f(b) f(a) GRASTA 2018, Berlin 19
Example of distinguishable configurations GRASTA 2018, Berlin 20
Property 1 • Let E = C1C2… Cxbe a factor of execution • If C1’ is indistinguishable from C1 by the automorphism f then – C1’ C2’ … Cx’ is a factor of execution such that ∀i∈ [1..x], Ciand Ci’ are indistinguishable by f i.e.∀p∈V, Multiplicity(Ci,p) = Multiplicity(Ci’,f(p)) GRASTA 2018, Berlin 21
Illustration of the property a a b a b c b c c d d e d e f e f f g g h g h i h i i f(i) f(h) f(g) f(i) f(h) f(g) f(i) f(h) f(g) f(f) f(e) f(d) f(f) f(e) f(d) f(f) f(e) f(d) f(c) f(b) f(a) f(c) f(b) f(a) f(c) f(b) f(a) GRASTA 2018, Berlin 22
Roadmap • Models and Exploration problems • The lower bound • Application: lower bound for the ring case • Proof of the result • Optimal algorithm for the ring case GRASTA 2018, Berlin 23
Applications of the Result • Assuming full visibility and weak multiplicity – In SSYNC, 4 robots are necessary and sufficient to probabilistically explore any ring of size n > 4 [TCS, 2013] – In ASYNC, 3 robots are necessary and sufficient to deterministically explore almost all grids [SSS, 2012] • (2*2 requires 4 and 3*3 requires 5) – In SSYNC, 4 robots are necessary and sufficient to probabilistically explore a torus of size l * L where 7 ≤ l ≤ L [NETYS, 2015] GRASTA 2018, Berlin 24
Applications of the Result • Assuming full visibility and weak multiplicity – In SSYNC, 4 robots are necessary and sufficient to probabilistically explore any ring of size n > 4 [TCS, 2013] – In ASYNC, 3 robots are necessary and sufficient to deterministically explore almost all grids [SSS, 2012] • (2*2 requires 4 and 3*3 requires 5) – In SSYNC, 4 robots are necessary and sufficient to probabilistically explore a torus of size l * L where 7 ≤ l ≤ L [NETYS, 2015] GRASTA 2018, Berlin 25
Detailed Application • In a ring with K=3 robots (K=1 and K=2 are trivial insufficient using our theorem) GRASTA 2018, Berlin 26
Detailed Application • In a ring with K=3 robots GRASTA 2018, Berlin 27
Detailed Application • In a ring with K=3 robots GRASTA 2018, Berlin 28
Detailed Application • In a ring with K=3 robots GRASTA 2018, Berlin 29
Detailed Application • In a ring with K=3 robots GRASTA 2018, Berlin 30
Detailed Application • In a ring with K=3 robots GRASTA 2018, Berlin 31
Detailed Application • In a ring with K=3 robots GRASTA 2018, Berlin 32
Detailed Application • In a ring with K=3 robots • For n > 4, 4 robots are necessary – N.b. this does not exclude the case K=3 and n=4 GRASTA 2018, Berlin 33
Roadmap • Models and Exploration problems • The lower bound • Application: lower bound for the ring case • Proof of the result • Optimal algorithm for the ring case GRASTA 2018, Berlin 34
Proof of the Result Theorem: deterministic) exploration with – K robots – on a graph of n>K nodes is possible only if • K>2 and there exists a set S of at least n-K+1 configurations such that: – any two different distinguishable, and – in every configuration in S, there is a tower of less than K robots. Terminating (probabilistic or configurations in S are GRASTA 2018, Berlin 35
Proof of the Result • Let P be any exploration protocol for K robots on a ring of n > K nodes GRASTA 2018, Berlin 36
Proof of the Result • Let P be any exploration protocol for K robots on a ring of n > K nodes • As n > K and robots are oblivious, any terminal configuration should be distinguishable from any initial (towerless) configuration, hence: Remark 1: Any terminal configuration contains a tower. GRASTA 2018, Berlin 37
Proof of the Result Let S = C1C2… Cxbe a sequence of configurations MRS(S) is the maximal subsequence of S where no two consecutive configurations are identical Lemma 1: Let E = C1C2… Cxbe a sequential terminating execution of P. K > 2 and MRS(E) has at least n-K+1 configurations containing a tower of less than K robots. GRASTA 2018, Berlin 38
Proof of the Result • Let E = C1C2… Cxbe a sequential terminating execution of P GRASTA 2018, Berlin 39
Proof of the Result • Let E = C1C2… Cxbe a sequential terminating execution of P • By Remark 1, Cxcontains a tower and K>1 GRASTA 2018, Berlin 40
Proof of the Result • Let E = C1C2… Cxbe a sequential terminating execution of P • By Remark 1, Cxcontains a tower and K>1 • Let E’ = Ci… Cxbe the suffix of E such that – Ciis the only towerless configuration of E’ GRASTA 2018, Berlin 41
Proof of the Result • Let E = C1C2… Cxbe a sequential terminating execution of P • By Remark 1, Cxcontains a tower and K>1 • Let E’ = Ci… Cxbe the suffix of E such that – Ciis the only towerless configuration of E’ • E’ is a sequential terminating execution of P GRASTA 2018, Berlin 42
Proof of the Result • Let E = C1C2… Cxbe a sequential terminating execution of P • By Remark 1, Cxcontains a tower and K>1 • Let E’ = Ci… Cxbe the suffix of E such that – Ciis the only towerless configuration of E’ • E’ is a sequential terminating execution of P • MRS(E’) is a sequential terminating execution of P too GRASTA 2018, Berlin 43
Proof of the Result • For each step C→C’ of MRS(E’), 3 cases: a) C is towerless (the first step). C’ contains a tower and no new node is visited in this step b) C contains a tower and C’ contains a tower of K robots. No new node is visited in this step c) C contains a tower and C’ contains a tower of less than K robots. At most 1 new node is visited in this step If K=2, case c) do not exist: except the K initially visited nodes, no other node can be visited If K>2: Initially, K nodes are visited. In case a), which appears exactly once, C’ contains a tower of less than K robots. Case c) should appear at least n-K times. Hence, MRS(E’), and so MRS(E), has at least n-K+1 configurations containing a tower of less than K robots. GRASTA 2018, Berlin 44
Proof of the Result Lemma 2: Let E = C1C2… Cxbe a sequential terminating execution of P. K > 2 and MRS(E) has at least n-K+1 configurations containing a tower of less than K robots and any two of them are distinguishable. By the contradiction: – at least n-K+1 configurations containing a tower of less than K robots – But less than n-K+1 of them are pairwise distinguisable By inductively applying Property 1, we can construct a sequential terminating execution E’ from E that has less than n-K+1 configurations containing a tower of less than K robots, contradicting Lemma 1 GRASTA 2018, Berlin 45
Proof of the Result • Assume Ciand Cj – contains a tower of less than K robots and – are indistinguishable by the automorphism f in C1… Ci… CjCj+1… Cx • C1… CiCj+1’ … Cx’ is a terminating execution of P where ∀y∈ [j+1..x], Cyand Cy’ are indistinguishable by f i.e.∀p∈V, Multiplicity(Cy,p) = Multiplicity(Cy’,f(p)) • We removed Ci GRASTA 2018, Berlin 46
Back to the case K=3 and n=4 • Our result only deals with of sequential solutions – (a particular case of the distributed scheduler) • So, the result on the ring does not exclude sequential solution for K=3 and n=4 GRASTA 2018, Berlin 47
Sequential solution for K=3 and n=4 GRASTA 2018, Berlin 48
Sequential solution for K=3 and n=4 • However, no distributed solution – (Combinatorial study) GRASTA 2018, Berlin 49
Roadmap • Models and Exploration problems • The lower bound • Application: lower bound for the ring case • Proof of the result • Optimal algorithm for the ring case GRASTA 2018, Berlin 50