Project AURIONAI Use-Cases Research & Innovation

End-to-End Driving Policies with Reasoning

Driving models that map sensor input straight to a planned trajectory, increasingly with a language model that reasons about the scene first.

Tier Grades

feasible

Tier A: Frontier AI labs

feasible

NVIDIA trains Alpamayo-R1 at 0.5B to 7B parameters and Tesla trains FSD on its fleet.

not feasible

Tier B: Indonesian AI labs

not feasible

Building a driving VLA needs a curated reasoning dataset and on-vehicle testing that the sources show only for NVIDIA and Tesla.

stretch

Tier C: Institution units such as ITB

stretch

Open Alpamayo-R1 weights allow evaluation in simulation, but on-vehicle tests are not shown for institution units.

not feasible

Tier D: Organisations such as SEMAI (PoT, PoC, PoV)

not feasible

Road-vehicle autonomy is not a proof-of-concept scope; delivery work would stop at evaluation.

Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.

Anchors

  • ComputeAlpamayo-R1: scaled from 0.5B to 7B parameters on the Cosmos-Reason backbone; 99 ms on-vehicle latency.
  • DataAlpamayo-R1: Chain of Causation dataset built by auto-labelling plus human review; RL post-training.
  • Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 44 authors on the Alpamayo-R1 paper (arXiv 2511.00088).
  • Time horizonMulti-year.
  • Shelf-life signalsAlpamayo-R1 reports a 35% lower close-encounter rate in closed-loop simulation than a trajectory-only baseline.
  • Pipeline roleTier A trains and releases driving VLAs; Tier C evaluates open weights in simulation; Tier B and Tier D have no observed role.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.

Who and What

Exemplars
NVIDIAAlpamayo-R1TeslaTesla FSD v14
Observed actors
NVIDIANVIDIATeslaTesla
Linked models
TeslaTesla FSD v14drivingexemplarNVIDIAAlpamayo-R1drivingexemplarNVIDIAAlpamayo 1.5-10Bdrivingproduced byTeslaTesla World Simulatorworld modelused byStateTransformerdrivingproduced by

Papers

Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.

Sources

  1. arxiv.org/abs/2511.00088 https://arxiv.org/abs/2511.00088

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "End-to-End Driving Policies with Reasoning", https://aurion.semai.id/topics/end-to-end-driving-policies-with-reasoning/, records of October 2026.