Exploration by luminous robots
Myopic Look–Compute–Move robots with a light of a few colours must visit every node of a ring, grid or torus, and stop or keep visiting; how many robots and colours are needed?
Problem statement and measuresexpand
\(k\) anonymous robots, each with a light that shows one of \(c\) colours, are placed on an anonymous graph \(G=(V,E)\) (a ring, a finite grid or a torus) in an initial configuration designated by the algorithm (positions and colours). Each robot repeats Look–Compute–Move cycles. Look: it sees every node within distance \(\varphi\), with the multiset of colours of the robots on each (so it sees towers, nodes with several robots), and its own colour. Compute: a deterministic function of this snapshot, the same for all robots, gives a new colour and a destination, its own node or a neighbour; the robot sees no identifiers, no ports, and (without chirality) no orientation. Move: it moves there. A fair adversarial scheduler (FSYNC, SSYNC or ASYNC) activates the robots. Terminating exploration: every node is visited by some robot and eventually no robot is enabled. Perpetual exploration: every node is visited infinitely often. With finite \(\varphi\) no algorithm works from every initial configuration, so each result names its initial configurations; some papers also require exclusive executions (no two robots on one node or one edge) or opaque robots, recorded under each result. Measures: the number of robots \(k\) and the number of colours \(c\).
Measures: Robots, number of robots \(k\) needed; Colours, number of colours \(c\) of each light; \(c = 1\) means no usable light (oblivious robots).
Rings, FSYNC, perpetual, up to \(f\) crashes
verified\(f + 2\) robots with \(3f + 2\) colours explore a ring perpetually despite \(f\) crashes, which is optimal in robots.
Assumed (added on the last edge are outlined)
Not assumed
Formal definitions of this setting
- Graph dynamics: static graph
- \(G_t=G\) for all \(t\): the edge set never changes.
- Edge direction: undirected
- \(G=(V,E)\) with \(E\subseteq\binom V2\); every edge can be traversed in both directions.
- Network: ring
- An anonymous, unoriented ring of \(n\) nodes; \(n\) is unknown to the robots.
- Exploration goal: perpetual
- Every node is visited by at least one robot infinitely often; robots never stop.
- Lights (memory and messages): luminous
- LUMI (luminous robots): each robot has a light with \(c\) colours, set at the end of Compute, persistent across cycles, and visible to the robot itself and to every robot within distance \(\varphi\) Das et al. 2016. It is both memory and message. A Look returns, for every node within distance \(\varphi\), the multiset of colours of the robots on it.
- Scheduler: FSYNC
- FSYNC: in every round every robot executes one atomic Look–Compute–Move cycle.
- Visibility range: \(\varphi=1\)
- In a Look phase a robot at node \(v\) sees every node at distance at most \(\varphi = 1\) from \(v\) (in the graph metric), with the robots on it, their colours, and how many robots each node holds. It sees no node identifiers and, without chirality, no orientation.
- Chirality: no chirality
- Robots share no orientation: on a ring a robot cannot tell its two directions apart; on a grid or torus its snapshot is known only up to rotation and reflection, and the adversary may choose a different transformation in every cycle (self-inconsistent robots, as in Darwich et al. 2023 and Bramas et al. 2026).
- Faults: at most crash faults
- Up to \(f\) agents may stop forever at an adversarially chosen time; a crashed agent stays at its node (or vanishes, as stated per result) and never acts again.
- Identifiers: anonymous
- All agents run the same deterministic algorithm on the same input; no agent has a label. Formally the label function is constant: \(\mathrm{id}(a)=\mathrm{id}(b)\) for all agents \(a,b\).
What is known
- at least two robots must survive (fault-free bound of two robots)
- any \(\varphi\) and any number of colours
Results, in order of publication
Introductory and current best in robots. A crashed robot keeps its colour and cannot be told apart from a correct one.
Oblivious robots; one of three robots–colours trade-offs.
Logarithmically many colours.
Open questions
With \(f + 2\) robots under FSYNC, how many colours are necessary for crash-tolerant perpetual ring exploration? The best algorithm uses \(3f + 2\) Ooshita et al. 2025.
Notes
The trivial algorithm runs \(f + 1\) independent pairs with distinct colours (\(2f + 2\) robots, \(2f + 2\) colours); Ooshita et al. 2025 lowers the robots to \(f + 2\).
What carries over from neighbouring settings
Read off the order of assumptions, not checked case by case: results may carry extra assumptions (shown on their pages), and asynchronous time is measured in epochs.
Algorithms from weaker settings work here
Lower bounds from stronger settings hold here
Know a better result for this setting, or found a mistake? Suggest a result · Report an error. Accepted contributions are acknowledged.