LectureRechercheRadarCadre d'investissement
Connexion / Inscription
← Retour à la lecture
ActualitésSujet · 机器人泛化瓶颈Prise rapide1 min de lectureRédaction DisconfirmAI

DeepMind researcher says robot intelligence is still at GPT-2 level

A field lead's yardstick: locomotion is no longer the bottleneck, cross-embodiment generalization is — useful for reading humanoid robot news.

2026-10-03 21:00 GMT+8

Keerthana Gopalakrishnan, research lead for Gemini Robotics at Google DeepMind, said on the October 3 episode of the Cognitive Revolution podcast that robot intelligence still sits at GPT-2 level on a 1-to-6 "how many GPTs" scale.

Per the notes published on the episode page, she locates the bottleneck not in instruction-following but in cross-embodiment — a policy that works on only one robot body is not yet a generic brain. She also separates two curves: locomotion trains well in simulation, while sim-to-real still breaks down on manipulation tasks like cloth and friction.

The episode notes are AI-generated summaries, not a verbatim transcript; the audio remains the authoritative record.

Sources

  • cognitiverevolution.ai↗
Sujet

机器人泛化瓶颈

谷歌DeepMind研究员Keerthana指出,当前机器人智能仅相当于GPT-2水平,运动能力已非主要障碍,跨本体泛化才是核心瓶颈【背景】。这一判断与中银国际研报及宇树科技创始人王兴兴的观点一致:固定场景成功率近100%,但环境微变即导致…

Connexion / Inscription
LectureRechercheRadarCadre d'investissement

5 étapes2 piècesMis à jour 2026-10-05 18:33

Comment on en est arrivé là

  1. 2026-08-10中银国际发布研报,明确大模型泛化能力不足是人形机器人商业化的主要瓶颈,指出全球合规真实物理交互数据严重匮乏。news.qq.com ↗
  2. 2026-08-20宇树科技创始人王兴兴在世界机器人大会上表示,具身智能最大瓶颈是泛化能力不够,固定场景外成功率大幅下降。bjnews.com.cn ↗
  3. 2026-09-19Figure AI宣布Helix 2.5模型在30个未接触过的家庭环境中完成56%的家务任务,称预训练策略使零样本成功率从9%提升至56%。techtimes.com ↗

Ce qui reste non résolu

  • Figure AI声称的56%零样本成功率能否在其他厂商的非结构化场景中复现?
  • 随着数据积累,人形机器人何时能真正跨越‘最后几厘米’的精度瓶颈进入大规模家庭服务?
  • 资本市场是否会因泛化进展不及预期而进一步修正头部机器人企业的估值逻辑?

Également dans ce fil

  • DeepMind robotics lead says general-purpose robots remain far from practical use2026-10-04
Lire le sujet complet →
Continuer la recherche →