<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Che (Percy) Liu — Research notes</title><description>Che (Percy) Liu is an MBZUAI undergraduate and student researcher working on AI agents, game technology, and developer tools.</description><link>https://thecyper.github.io/</link><language>en-us</language><item><title>KIMODO 能做什么、不能做什么：Prompt、Constraint 与案例评测</title><link>https://thecyper.github.io/work/kimodo-capability-boundaries/</link><guid isPermaLink="true">https://thecyper.github.io/work/kimodo-capability-boundaries/</guid><description>从一组成功与失败案例中拆解 KIMODO 的真实能力：哪些动作 prior 稳定、Constraints 能修什么，以及模型没有场景、道具状态或跨 Prompt 记忆意味着什么。</description><pubDate>Tue, 28 Jul 2026 00:00:00 GMT</pubDate><language>zh-CN</language><category>text-to-motion</category><category>motion-authoring</category><category>constraints</category><category>character-animation</category></item><item><title>Robust Motion In-Between：从可复现基准到 UE 5.8 Runtime</title><link>https://thecyper.github.io/work/robust-motion-in-between/</link><guid isPermaLink="true">https://thecyper.github.io/work/robust-motion-in-between/</guid><description>一次有边界的 RMIB 复现：先在 LAFAN1 上完成 M6、单步 ONNX 与 NNE 对拍，再转向独立的 Manny-native 流式推理与 bake 路径。</description><pubDate>Mon, 06 Jul 2026 00:00:00 GMT</pubDate><language>zh-CN</language><category>motion-in-betweening</category><category>unreal-engine</category><category>onnx</category><category>character-animation</category></item><item><title>AI 动画技术版图：生成、补帧、检索、约束与 Runtime Control</title><link>https://thecyper.github.io/writing/ai-animation-technology-map/</link><guid isPermaLink="true">https://thecyper.github.io/writing/ai-animation-technology-map/</guid><description>先拆开问题，再选择模型：用任务、运动来源与控制闭环理解 Motion Matching、In-betweening、Diffusion、Motion Token 与 Physics RL。</description><pubDate>Fri, 03 Jul 2026 00:00:00 GMT</pubDate><language>zh-CN</language><category>AI Animation</category><category>Motion Generation</category><category>In-betweening</category><category>Unreal Engine</category></item><item><title>UE 5.8 AnimGen：从 1,312 段有效动画到 Runtime Pose Generation</title><link>https://thecyper.github.io/work/animgen-runtime-pose-generation/</link><guid isPermaLink="true">https://thecyper.github.io/work/animgen-runtime-pose-generation/</guid><description>一次 UE 5.8 Preview 单卡实验：过滤 additive 数据，训练 AutoEncoder 与两阶段 Controller，并在 AnimBP 的 PIE 循环中验证单角色 runtime pose generation。</description><pubDate>Fri, 29 May 2026 00:00:00 GMT</pubDate><language>zh-CN</language><category>AI Animation</category><category>Unreal Engine</category><category>AnimGen</category><category>Motion Generation</category><category>PyTorch</category></item><item><title>UE 5.8 官方 MCP：从 237 个工具到可验证工作流</title><link>https://thecyper.github.io/work/ue-58-mcp-reliability/</link><guid isPermaLink="true">https://thecyper.github.io/work/ue-58-mcp-reliability/</guid><description>在 2026 年 5 月的 UE 5.8 Preview 中，用 22 个核心 Toolset 的原子测试与业务任务，拆解 AI 操控 Editor 真正需要的 discover → plan → execute → verify 闭环。</description><pubDate>Fri, 22 May 2026 00:00:00 GMT</pubDate><language>zh-CN</language><category>Unreal Engine</category><category>MCP</category><category>Tool Use</category><category>Agent Reliability</category></item></channel></rss>