LADA: Low-Cost Driving Annotations via Latent Action Models
abursuc · x · 2026-09-17
Presented at the #ssad2026 workshop, LADA (Latent Action Driving Annotations) is a low-cost strategy for generating language annotations for driving data. Built on latent action models (LAMs), it learns to produce latent actions from only a few observation-language pairs, imitating the function of language to annotate driving datasets cheaply. The post also includes a primer on LAMs.
More from Research
- Thinking With Looped Flows Hits 58.8% on ARC-AGI-1, Joining the Looped-Flow Reasoning Wave — alec_helbling · 2026-09-17
- Flow Reasoning Models Turn Flow Models Into Recurrent Reasoners, Hitting 99.5% on Sudoku-Extreme — alec_helbling · 2026-09-17
- FRM Matches EqR on Sudoku-Extreme With 44x Fewer FLOPs — alec_helbling · 2026-09-17
- Flow Reasoning Models: turning flow models into recurrent reasoners without BPTT — alec_helbling · 2026-09-17
- Gen-COMPAS: diffusion model proposes, physics decides for rare molecular transitions — bravo_abad · 2026-09-17
- epidemik 0.2.0: Define Epidemic Models as Strings, Get R0 and Diagrams for Free — bgoncalves · 2026-09-17