Research & Development
Pace delineation: distributed, stop-free coordination of vehicles whose paths share a conflict point.
The pace-delineation protocol
Wherever two trajectories cross — an intersection, a highway on-ramp, a warehouse aisle junction — exactly one vehicle may occupy the conflict point at a time. Conventional control serializes access by stopping traffic (signals, stop lines, AGV traffic managers that halt robots at junctions). JIBE instead separates vehicles in time: agents negotiate their arrival instants while still upstream and make small, bounded speed adjustments so that occupancy windows at the conflict point never overlap. The result is the same safety guarantee as a stop-based scheme, with the kinetic energy and schedule of every vehicle largely preserved.
- 1.State broadcast. Each agent periodically announces position, velocity, and intended path over V2V. No central controller is required; any agent (or roadside unit, where present) can run the same allocation rule and reach the same answer.
- 2.Conflict detection. Agents project arrival times at shared conflict points. Two occupancy windows closer than the required headway h constitute a conflict.
- 3.Slot allocation. The joining agent evaluates the feasible arrival slots between existing arrivals and claims the one minimizing total pace deviation across all affected vehicles. The claim is explicit and acknowledged.
- 4.Pace adjustment. Required speed changes are bounded by comfort limits and a hard speed floor, and are applied as far upstream as the communication horizon allows — the earlier the negotiation, the smaller the adjustment.
- 5.Traversal and release. After crossing, the slot is released and adjusted agents resume nominal speed. Arrival error versus the committed time t* is the protocol's precision metric.
Design properties
- Auditable: every speed change traces to a logged message exchange.
- Decentralized: allocation is deterministic given the broadcast state, so agents agree without a coordinator.
- Bounded actuation: adjustments respect comfort and stability limits; the protocol degrades to holding upstream rather than emergency braking.
- Incremental deployment: unequipped vehicles are treated as non-negotiable constraints; benefit scales with adoption.
Transport assumptions
- Periodic state broadcast at 10 Hz over DSRC or C-V2X (PC5) on roads; Wi-Fi/private 5G indoors.
- Negotiation completes in a handful of messages, so it tolerates the latency and loss typical of these links.
- Missed acknowledgements fall back to renegotiation or upstream hold — silence is never interpreted as consent.
Message set
The negotiation reduces to a small, explicit vocabulary. These are the same messages traced in the demo log below.
| Message | Direction | Payload | Semantics |
|---|---|---|---|
| STATE / MRG_REQ | agent → broadcast | id, position, velocity, intended path, ETA at conflict point | Periodic state announcement; MRG_REQ marks entry into an active negotiation. |
| SLOT_CLAIM | joining agent → trailing agent | t* (committed arrival time), predecessor id | Claims the arrival slot selected by the allocation rule (minimum total pace deviation). |
| YIELD_ACK / CLEAR_ACK | trailing agent → joining agent | Δv (bounded speed adjustment, possibly zero) | Accepts the claim. Δv is limited by comfort bounds and a speed floor — never a stop. |
| MERGE_DONE / RELEASE | joining agent → broadcast | arrival error vs. t* | Slot is released; adjusted agents resume nominal speed. |
| RETRY | joining agent → broadcast | updated ETA | No feasible slot under current constraints; agent holds upstream and renegotiates. |
Demo: cooperative ramp merge
When a ramp vehicle crosses the V2V horizon it broadcasts MRG_REQ with its ETA at the merge point, then claims an arrival slot between two mainline vehicles (SLOT_CLAIM). If the trailing vehicle must arrive later, it acknowledges a bounded speed reduction (YIELD_ACK) — a pace adjustment, never a stop. Vehicle IDs in the message log match the labels on the canvas, and every value shown (ETA, t*, Δv) is read from the live negotiation.
Shortening the V2V horizon or the headway shows why early negotiation matters: with less distance left to re-pace, the protocol must either ask the mainline for larger Δv or hold the ramp vehicle.
Application domains
The protocol is domain-agnostic: it needs agents with known kinematics, a shared conflict point, and a communication channel. The constraints differ; the negotiation does not.
Road traffic
Ramp merges and unsignalized intersections for connected and autonomous vehicles. The hard problems are mixed traffic — human-driven vehicles participate only as predicted, non-negotiable constraints — and the latency/reliability envelope of DSRC and C-V2X links. Comfort bounds on Δv keep adjustments imperceptible to passengers.
Warehouse AGV / AMR fleets
Aisle crossings and dock approaches, where today's fleet managers typically serialize junctions by halting robots. Controlled environments make this the nearest-term deployment: full equipage, reliable local networking, and known vehicle dynamics. Fewer stop/start cycles also reduce energy draw and drive wear, which we quantify per site rather than as a headline number.
Low-altitude airspace
Corridor crossings and vertiport approaches for UAS/UAM traffic. The same slot allocation applies with the conflict point generalized to a 4D volume (position + time); pace adjustment is the natural control input for aircraft, which cannot stop and wait.
The same structure applies to port terminals, mining haul roads, and rail-yard shunting — any setting with schedulable agents and intersecting paths.
Open problems we are working on
Multi-vehicle coordination at conflict points has no deployed standard today. These are the questions that decide whether a protocol like this one is deployable, and they are where our current research effort goes.
- Mixed traffic. Guaranteeing safety when some agents never negotiate: conservative prediction envelopes for unequipped vehicles, and how quickly benefit accrues with partial adoption.
- Degraded communication. Behavior under packet loss, latency spikes, and network partitions — proving that the fallback (hold upstream, renegotiate) is safe and live under realistic channel models.
- ETA uncertainty. Slot allocation with noisy state estimates: sizing headway h against localization and actuation error rather than fixing it by convention.
- Formal verification. Model-checking the negotiation state machine for deadlock-freedom and collision-freedom, so the safety argument does not rest on simulation alone.
- Strategic behavior. Allocation rules that remain efficient when agents misreport state to win earlier slots — incentive compatibility for open road deployments.