One-shot imitation performance remains flat despite model scale
animesh_garg · x · 2026-08-30
Animesh Garg compares 2020 and 2026 robotics techniques: while modern generalist models (like GEN-1.5 and Skild) achieve "emergent" physical in-context learning directly from video prompts, the performance rates for one-shot imitation remain roughly the same as those achieved in 2020 using learned Neural Task Graphs (NTG). He argues that task specification is effectively a code generation task and predicts high-performance systems will rely on code generation (e.g., ProgPrompt) using atomic skills.
Related event: Robot One-Shot Imitation: Bigger Models, Little Real Progress(2 posts)→
More from Embodied
- Autonomous launches AI companion robot Lamp for $499 with skill store — dee_hw · 2026-08-30
- Hyundai Unveils MobED, an AI Robot Built for Rough Terrain — TinfoilTricorn · 2026-08-30
- Autonomous cars need 4 separate pods for private carpooling — Darpinian · 2026-08-30
- ESPectre Enables Camera-Free Motion Detection via Wi-Fi CSI — tom_doerr · 2026-08-30
- Anthropic Launches MHS Standard to Enable AI Agents to Control Lab and Manufacturing Hardware — 创业邦 · 2026-08-30
- FrankenRobots on Patrol video — doodlestein · 2026-08-30