机器人如何在不确定的真实环境里理解任务、识别可操作对象，并稳定地完成动作？
How can a robot understand a task, identify actionable objects, and execute reliably in an uncertain physical environment?
DIRECTION DOSSIER · 01 / ONGOING
具身机器人Embodied
操作Manipulation从看见，到理解，再到动手From seeing and understanding to acting
我关注机器人如何把视觉、语言与身体状态连接起来，形成可执行、可复盘的操作行为。这是一份正在生长的研究方向档案，不把尚未完成的内容包装成“成品”。
I am interested in how robots connect vision, language, and bodily state into actions that can be executed and reviewed. This is a growing direction dossier, not a finished project presented as one.
状态 · 方向研究与实验准备中STATUS · DIRECTION STUDY IN PROGRESS
01 / OVERVIEW
WHAT THIS DOSSIER HOLDS
先把问题说清楚。Frame the problem first.
当前聚焦机器人操作、具身智能与多模态感知之间的接口，先建立清晰的系统链路，再逐步补充实验。
The current focus is the interface between manipulation, embodied intelligence, and multimodal perception—first clarifying the system, then adding experiments.
02 / ROUTE
WORKING PIPELINE
一条可验证的操作链路。A testable manipulation pipeline.
- 01感知环境Perceive
整合视觉、点云与任务描述，建立对象和空间关系。Combine vision, point clouds, and task descriptions into object and spatial relationships.
- 02形成策略Plan
把目标转化为动作序列，并在仿真或受控场景中先行验证。Translate goals into action sequences and test them first in simulation or controlled scenes.
- 03执行与反馈Act & Review
关注执行稳定性、失败恢复，以及每次实验能否沉淀出下一轮改进。Focus on execution stability, recovery from failure, and lessons that shape the next iteration.
03 / EVIDENCE
TO BE CONNECTED
接下来补真实证据。Real evidence comes next.
实验场景Experiment scene
任务目标、对象与环境约束。Task, objects, and environmental constraints.
技术路线Technical route
模型、工具、系统结构与个人角色。Models, tools, system structure, and personal role.
结果与失败Results & failures
数据、视频、问题与下一轮迭代。Data, video, failure modes, and the next iteration.
04 / NEXT
OPEN LOG
档案会随着项目继续生长。This dossier will grow with the work.
下一步是把具体实验、项目职责、过程图与可复现结果接入这里。主页只给出入口，这个独立网址负责完整叙事。
The next step is to connect concrete experiments, responsibilities, process imagery, and reproducible results. The homepage provides the entry; this route carries the full story.
← 回到作品列表← BACK TO SELECTED WORK