AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

SWE-MiniSandbox: Container-Free Reinforcement Learning for Building Software Engineering Agents

Danlong Yuan, Wei Wu, Zhengren Wang, Xueliang Zhao, Huishuai Zhang, Dongyan Zhao

ICML 2026 · Reinforcement learning › Visual and representation RL · approach: Probabilistic and autonomous RL · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use MiniSandbox instead of Docker containersSWE-MiniSandbox · RL rollout, 3Benv prepare time (s): 88.86 -> 23.62Improved
use MiniSandbox instead of Docker containersSWE-MiniSandbox · RL rollout, 3Bavg rollout time (s): 367.33 -> 272.71Improved
use MiniSandbox instead of Docker containersSWE-MiniSandbox · SWE-bench Verified, 3B after RLresolve rate (%): 8.6 -> 9.2No clear effect
use MiniSandbox instead of Docker containersSWE-MiniSandbox · SWE-bench Verified, 7B after RLresolve rate (%): 12.4 -> 11.8No clear effect
use MiniSandbox instead of Docker containersSWE-MiniSandbox · SWE-bench Verified, SWE-Agent-7B after RLresolve rate (%): 16.4 -> 16.8No clear effect
use MiniSandbox instead of image-based storageSWE-MiniSandbox · SWE-smith env cacheenvironment storage: 295 GB -> 13.5 GBImproved
use MiniSandbox instead of one-to-one imagesSWE-MiniSandbox · SWE-bench Verified env cacheenvironment storage: 605 GB -> 89 GBImproved
use MiniSandbox instead of Docker containersSWE-MiniSandbox · reward computationreward time (s): 14.89 -> 9.96Improved
use MiniSandbox instead of Docker containersSWE-MiniSandbox · environment communication timeoutstimeouts per instance: 0.93 -> 0.39Improved
scale rollout count to 16 (256 parallel envs, 1 node) (ablation)SWE-MiniSandbox · MiniSandbox vs Dockerenv prepare time (s): MiniSandbox 20.09 -> 116.25 vs Docker 87.45 -> 123.79Hurt
scale from 1 to 2 nodes at 256 envsSWE-MiniSandbox · MiniSandboxenv prepare time (s): 116.25 -> 20.72Improved
scale from 1 to 2 nodes at 256 envsSWE-MiniSandbox · Dockerenv prepare time (s): 123.79 -> 113.67No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy