v0.2.0 Deferred Work — Future Swarm Implementation

This document records every swarm-related feature or behaviour discussed during v0.2.0 planning that is intentionally left out of this release. It is the backlog seed for later milestones. Do not treat absence from the v0.2.0 implementation checklist as “forgotten.”

Related:

• In scope: docs/research/swarm_coordination.md and docs/swarm.md
• Research: docs/research/swarm_coordination.md
• Formation guide: docs/formations.md

Suggested roadmap homes are indicative, not commitments. Revisit when that milestone is planned.

1. Behavioral flocking (decentralized local rules)

Discussed: Swarms can hold structure without a central leader by relying on local interactions (Reynolds-style / boids).

Future module sketch: Drone.Swarm.Behavior (or Drone.Flocking)

• Periodic tick (simulator-first)
• Each vehicle reads neighbor telemetry within a radius
• Emits soft relative move / rotate suggestions (still through Safety)
• No central leader required for structure maintenance

Why deferred: Needs a neighbor sensing loop, timing model, and non- deterministic-friendly test strategy. v0.2.0 ships geometric planners, not emergent flocking.

Likely home: after swarm primitives stabilize; possibly with richer safety / observability work.

2. Live / dynamic collision avoidance

Discussed: Reject or reshape commands that would collide with other swarm members during flight.

Future sketch:

• Vehicle Safety gains optional neighbors: [%{name, x, y, z}, ...]
• Swarm periodically publishes member poses to members (or Safety queries swarm)
• On predicted breach of min_separation_cm, reject move or force hover

Why deferred: Changes the Safety contract and requires a pose distribution path. Easy to get wrong on real hardware without global localization.

3. Continuous / closed-loop formation control

Discussed: Hold Front/Vee/etc. over time while the group translates or turns (PID / leader-follower / virtual structure).

Why deferred: v0.2.0 formations are command generators, not pilots. Closed-loop needs reliable pose and a control period.

4. Leader-based runtime dependency and recovery

Discussed: Formations relative to an explicit leader; what if the leader fails?

v0.2.0 decision:

• Default reference = centroid or configured swarm origin, not a living leader
• Optional leader: means plan-time pose snapshot only
• Missing leader at plan time → hard error (:leader_unavailable)
• Leader command failure during run → same as any member (:fail_fast)
• No auto-undo of followers

Deferred behaviours:

Why deferred: Recovery policy is product-sensitive and unsafe to guess. Callers use explicit land/1 or emergency/1 in v0.2.0.

5. Alternate swarm failure policies

Discussed: :fail_fast vs :best_effort vs :all_or_nothing.

Why deferred: :all_or_nothing implies automatic dangerous commands. Must be explicit, tested, and documented before enabling.

6. Async / concurrent fan-out

Discussed: Task.async_stream for faster real-hardware coordination; Mission.concurrent/2 for parallel per-drone missions.

Why deferred: Concurrency complicates tests and partial-failure semantics. Sim-first milestone prioritizes determinism.

7. Automatic retry of swarm / movement steps

Discussed: Retry failed formation legs or movement commands.

v0.2.0: No automatic retry of dangerous movement commands (existing core principle). Swarm does not retry failed member steps.

Future (only with explicit enablement):

• Query retries remain separately allowable (already a v0.1.0 principle)

• Optional swarm retry: [max: n, only: [:query | ...]] — never default-on for moves

• Human-in-the-loop replay of failed member missions

8. Multi-node / distributed swarms

Discussed: Swarm members on multiple BEAM nodes.

v0.2.0: Single-node only.

Future:

• Distributed Registry or explicit node-qualified member refs
• Partition handling (split-brain is hazardous for physical actuators)
• Cross-node emergency propagation

Likely home: post-v1.0 research; treat as advanced/ops concern.

9. Nested swarm supervisors / per-swarm vehicle ownership

Discussed: DynamicSupervisor per swarm that owns vehicle children; richer restart strategies for the group.

v0.2.0: Vehicles stay under Drone.Supervisor; swarm is a sibling coordinator.

Future:

• Optional “swarm owns vehicles” mode for classroom demos
• Group restart policies (restart all members if one crashes — usually wrong for hardware, maybe useful for sim)

10. Real-hardware absolute formation flight (Tello)

Discussed: Running geometric formations on real Tello / Tello EDU.

v0.2.0: Simulator-first. Docs warn about Wi-Fi AP vs EDU station mode and lack of global pose. No claim of closed-loop choreography on bare Tello.

Future:

• Tello EDU station-mode multi-drone connectivity guide
• External localization (UWB, mocap, vision) feeding vehicle state
• Hardware-in-the-loop tests behind explicit tags / opt-in CI
• Honest accuracy bounds for open-loop slot flying on consumer drones

11. Adapter / platform swarm work beyond Sim

Swarm and Formation modules must stay adapter-agnostic in v0.2.0 so these can plug in later.

12. Sensing, video, and surveillance semantics

Discussed: Circular/Diamond layouts for 360° surveillance and defense; Vee for aerodynamic drag savings.

v0.2.0: Geometry only. No sensor tasking, no video sync, no aero model.

Deferred:

13. Formation catalog extensions and motion relative to heading

v0.2.0 ships geometric planners for classic shapes (see design plan): :front, :column, :vee, :diamond, :echelon, :circle, plus aliases :shoulder_pair / optional :grid.

Still deferred beyond those planners:

Heading-relative layout (heading_deg) is in scope for v0.2.0 planners so Front/Column/Vee/Echelon are meaningful; continuous heading tracking while moving is not.

14. Mission orchestration extras

15. Observability and operator tooling

16. Membership dynamics

Tracking rule

When a deferred item above is scheduled:

1. Move or copy its section into that milestone’s research/design docs
2. Strike or annotate it here with Implemented in vX.Y.Z
3. Do not implement it opportunistically inside v0.2.0 without updating the review checklist

Summary table

See also

• Guides: Swarms, Formations
• Research: Swarm Coordination
• Further reading: Swarming, formations, and strategies