---
title: "ADEPT：通过预训练与后训练强化学习加速灵巧操作"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.19182v1"
permalink: "https://feed.astralor.com/items/itm_0bc4d907efc2fb92cbf934d3"
published: "2026-08-19T17:55:06.000Z"
score: 78
channel: "ai"
tags: ["ADEPT","Flexiv-Sharpa","Kuka-Allegro","仿真到现实迁移","强化学习","机器人灵巧操作"]
---

# ADEPT：通过预训练与后训练强化学习加速灵巧操作

## AI 点评

> 这份工作把大模型预训练思路引入灵巧操作，最值得关注的是它解决了朴素微调导致能力遗忘的问题，并且用触觉+视觉在多指手上实现了zero-shot迁移。对做机器人学习、RL算法或具身智能的开发者来说，ADEPT的稳定后训练配方和Geometric Fabric设计都有直接参考价值。

## 速览

ADEPT是一种大规模强化学习框架，旨在通过预训练和后训练实现跨高自由度机器人形态的仿真到现实灵巧操作迁移。该框架先在通用物体重定位任务上预训练灵巧策略，再将其作为先验用于下游任务的后训练，从而避免从零学习并加速新技能获取。研究提出稳定后训练方案（行为克隆蒸馏、评论家热身、保守策略更新）以及联合空间几何织构，成功在23自由度Kuka-Allegro和29自由度Flexiv-Sharpa上实现零样本仿真到现实迁移，并以人类级速度解决长时域任务。

## 关键点

- 提出ADEPT框架，通过预训练+后训练加速多指机器人灵巧操作学习，避免每个新任务重复学习基础技能。
- 设计稳定后训练配方（行为克隆蒸馏、批评者热身、保守策略更新），解决朴素RL微调导致预训练能力退化的问题。
- 引入联合空间几何织构（Geometric Fabric），在RL策略与机器人之间提供安全且充分利用运动学灵巧性的中介层。
- 在23自由度Kuka-Allegro和29自由度Flexiv-Sharpa上实现零样本仿真到现实迁移，可从复杂初始状态以人类级速度完成长时域任务。

## 全文

We introduce Accelerating Dexterity via Pre-Training (ADEPT), a large-scale reinforcement learning (RL) framework for learning sim-to-real transferable dexterity across high degree-of-freedom (DoF) robot embodiments that can solve long-horizon tasks directly from raw visuo-tactile perception. ADEPT pretrains a dexterous policy on a generic object reposing task, then post-trains downstream policies with this pretrained behavior as a prior. ADEPT enables learning new behaviors that are otherwise difficult to discover from scratch on multi-fingered robots and avoids learning the same set of skills over again for every new downstream task. The pretrained policy zero-shots the reposing phase of downstream tasks, but naïve RL fine-tuning rapidly degrades this capability during transfer. We address this with a stable post-training recipe combining behavior-cloning distillation, critic warm-up, and conservative on-policy updates. To safely exploit the full kinematic dexterity, we introduce a joint-space Geometric Fabric that mediates between the RL policy and the robot. We distill post-trained teachers into perceptive students that zero-shot sim-to-real transfer on two embodiments: a 23 DoF Kuka-Allegro with two RGB cameras, and a 29 DoF Flexiv-Sharpa with two RGB cameras and five vision-based tactile sensors, and can solve long-horizon tasks from challenging initial states with dexterity at human-level speed.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_0bc4d907efc2fb92cbf934d3