[{"content":"Overview This direction explores how 3D Gaussian Splatting scene representations can support instance navigation and visual localization. The page is intentionally high-level. It does not disclose information beyond the public boundary of the 3DGS/patent-related work associated with Tsinghua\u0026rsquo;s vehicle-engineering context.\nProblem A representation designed for rendering is not automatically a scene model suitable for robot decisions. The research question is how view representation, spatial queries, instance semantics, and localization requirements can be placed within an evaluable system boundary.\nSystem Architecture The public boundary contains only four layers: scene capture and representation → queryable 3D information → navigation/localization interface → task-level evaluation. Pre-filing algorithm steps, parameters, data, and implementation details are withheld.\nResults Current status: patent disclosure in preparation. This site does not claim a granted patent, a published paper, or performance results without public evidence.\nLimitations This is not a technical disclosure. The core approach, data, comparisons, and contribution boundary will be updated only when publication is permitted and evidence has been reviewed.\n","permalink":"https://functionhx.github.io/en/research/3dgs-navigation-localization/","summary":"Only the research question, system boundary, and status are public; core pre-disclosure technical details are withheld.","title":"3DGS for Navigation and Visual Localization"},{"content":"Overview This page records Formula Student Driverless experience at a public, high level. It does not attribute a team\u0026rsquo;s results to one person or publish unverified competition rankings or performance metrics.\nProblem A driverless Formula Student system must coordinate perception, state estimation, planning, control, and vehicle interfaces under finite compute and real track constraints. The engineering challenge includes time synchronization, fault isolation, observability, and repeatable testing—not only individual algorithms.\nSystem Architecture The public record follows a clear boundary: sensors and vehicle interfaces → state estimation and scene understanding → trajectory planning → control execution → logging and safety monitoring. Verified personal contributions and evidence can be added within this structure.\nLimitations The team name, specific role, season results, and quantitative performance have not been published because they were not supplied for verification.\n","permalink":"https://functionhx.github.io/en/projects/formula-student-driverless/","summary":"A project record in autonomous-system integration and track engineering.","title":"Formula Student Driverless"},{"content":"Overview The currently public fact is: 20 upstream pull requests have been merged. Representative ecosystems include vLLM, OpenRLHF, NVIDIA CCCL, CycloneDDS, ROS 2, and OpenCV.\nContribution Areas AI Infrastructure Robotics Middleware Developer Tools Evidence Policy Repository names are not used to infer specific contributions, and missing links are not replaced with invented pull-request numbers, line counts, or impact claims. Each contribution will be added after its URL, merge status, and authorship are verified.\n","permalink":"https://functionhx.github.io/en/open-source/merged-pull-requests/","summary":"Across AI infrastructure, robotics middleware, and developer tools; individual links await verification.","title":"Merged Upstream Pull Requests"},{"content":"","permalink":"https://functionhx.github.io/en/about/biography/","summary":"","title":"Biography"},{"content":"Overview This page records designated-target following work on Unitree G1. Its public scope focuses on the system problem and interface boundaries without inferring an undisclosed model, sensor configuration, or role.\nProblem Target following requires a stable loop across target selection, persistent perception, occlusion recovery, motion commands, and safety constraints. A real robot must also handle communication latency, state transitions, and fail-safe behavior.\nSystem Architecture Verified material will follow this structure: target specification → perception and tracking → relative-state estimation → behavior and motion interface → safety monitoring.\nLimitations Unverified hardware configurations, model metrics, scene success rates, and team responsibilities are not published.\n","permalink":"https://functionhx.github.io/en/projects/unitree-g1-target-following/","summary":"A real-robot system record organized around target selection, persistent tracking, motion interfaces, and safety boundaries.","title":"Unitree G1 Target Following"},{"content":"Overview This direction studies how world models can assist editable road-scene generation so that scene changes follow explicit conditions and can support later analysis or system validation.\nProblem Road-scene generation requires more than visual plausibility. It also raises questions of controllable editing, spatiotemporal consistency, condition preservation, and interpretable evaluation. Visual examples alone are not treated as proof of system effectiveness.\nSystem Architecture Only the research boundary is public: scene conditions → representation and generation → controllable editing → consistency and task evaluation. Unpublished implementation details, data, and results are withheld.\nResults Current status: manuscript in preparation. This does not imply that a paper has been submitted, accepted, or released.\nLimitations Unverified metrics, comparative conclusions, and core technical details will not be published before the manuscript and public materials are ready.\n","permalink":"https://functionhx.github.io/en/research/editable-road-scene-generation/","summary":"While the manuscript is in preparation, only the problem, evaluation boundary, and limitations are public.","title":"World-Model-Assisted Editable Road-Scene Generation"},{"content":"","permalink":"https://functionhx.github.io/en/about/experience-timeline/","summary":"","title":"Experience Timeline"},{"content":"","permalink":"https://functionhx.github.io/en/about/resume/","summary":"","title":"Résumé"},{"content":"Overview This page records RoboMaster sentinel navigation and localization experience and keeps it distinct from the radar-station multi-sensor fusion work. It does not infer an undisclosed season, team, or personal role.\nProblem Navigation in a dynamic adversarial environment must continuously handle localization drift, moving obstacles, local traversability, path replanning, and recovery after system faults.\nSystem Architecture The public structure follows sensor input → state estimation → map and cost representation → global/local planning → chassis interface → runtime monitoring.\nLimitations Competition results, navigation metrics, hardware parameters, and contribution percentages are not shown without verification.\n","permalink":"https://functionhx.github.io/en/projects/robomaster-sentinel-navigation/","summary":"Localization, mapping, planning, and recovery under dynamic adversarial conditions.","title":"RoboMaster Sentinel Navigation"},{"content":"Overview This page records deployment work for VLA / VLN capabilities on Unitree Go2. It does not equate a research model\u0026rsquo;s general capability with a robot system that is ready to operate.\nProblem Moving from a model to a real robot requires explicit observation and command formats, inference latency, action constraints, state machines, communication-failure handling, and human takeover boundaries. Deployment reliability is a property of the model, system, and safety policy together.\nSystem Architecture The public boundary is sensing and task input → model-inference adapter → behavior/action interface → robot execution → telemetry and safety monitoring.\nLimitations The model version, scenario success rate, dataset, compute hardware, and detailed role are not published because they were not supplied for verification.\n","permalink":"https://functionhx.github.io/en/projects/unitree-go2-vln-deployment/","summary":"Model I/O, robot interfaces, the inference path, and safe operating boundaries.","title":"Unitree Go2 VLA / VLN Deployment"},{"content":"","permalink":"https://functionhx.github.io/en/about/contact/","summary":"","title":"Contact"},{"content":"Overview This page records Batch-LIO-related experience. Because a paper, repository, version, and personal contribution statement have not yet been supplied, it makes no additional claim about algorithm ownership or upstream results.\nProblem LiDAR–inertial odometry must balance asynchronous sensor data, motion distortion, state estimation, map representation, and compute cost while making failures explainable through repeatable data and logs.\nSystem Architecture Future evidence will follow LiDAR/IMU input → timing and motion processing → state estimation → map/registration interface → trajectory and diagnostic output.\nLimitations Unverified accuracy, speed, dataset results, code links, and contribution percentages are not published.\n","permalink":"https://functionhx.github.io/en/projects/batch-lio/","summary":"A project page structured around LiDAR–IMU data, state estimation, compute boundaries, and reproducible evidence.","title":"Batch-LIO"},{"content":" This page preserves the structure of the former Critiques section. It does not evaluate a real paper or technical paradigm.\nClaim State the subject\u0026rsquo;s central claim accurately before evaluating it.\nAnalysis Inspect gaps between the method, assumptions, evidence, and conclusion.\nEvidence / Counterexamples Record verifiable data, citations, reproductions, or counterexamples.\nVerdict Give a provisional judgment proportional to the evidence and state its limitations.\n","permalink":"https://functionhx.github.io/en/notes/critiques/example/","summary":"\u003cblockquote\u003e\n\u003cp\u003eThis page preserves the structure of the former Critiques section. It does not evaluate a real paper or technical paradigm.\u003c/p\u003e\n\u003c/blockquote\u003e\n\u003ch2 id=\"claim\"\u003eClaim\u003c/h2\u003e\n\u003cp\u003eState the subject\u0026rsquo;s central claim accurately before evaluating it.\u003c/p\u003e\n\u003ch2 id=\"analysis\"\u003eAnalysis\u003c/h2\u003e\n\u003cp\u003eInspect gaps between the method, assumptions, evidence, and conclusion.\u003c/p\u003e\n\u003ch2 id=\"evidence--counterexamples\"\u003eEvidence / Counterexamples\u003c/h2\u003e\n\u003cp\u003eRecord verifiable data, citations, reproductions, or counterexamples.\u003c/p\u003e\n\u003ch2 id=\"verdict\"\u003eVerdict\u003c/h2\u003e\n\u003cp\u003eGive a provisional judgment proportional to the evidence and state its limitations.\u003c/p\u003e","title":"[Archived Example] Anatomy of a Critique"},{"content":"This is the first post from the site\u0026rsquo;s initial setup. It records the original choice of Hugo and PaperMod and remains as migration history; it no longer represents the current information architecture or deployment setup.\n","permalink":"https://functionhx.github.io/en/notes/hello-world/","summary":"\u003cp\u003eThis is the first post from the site\u0026rsquo;s initial setup. It records the original choice of Hugo and PaperMod and remains as migration history; it no longer represents the current information architecture or deployment setup.\u003c/p\u003e","title":"Hello, World"},{"content":"","permalink":"https://functionhx.github.io/en/fx/","summary":"","title":"Effects Lab"}]