Rename public branding to Navigation Computing Optimization

Replace Compute Optimization / Compute Skill wording in titles, docs,
benchmark labels, and package tags so the published identity matches
navigation computing optimization without changing import or CLI paths.

Author: i6bimua

.github/ISSUE_TEMPLATE/feature.yml

@@ -1,5 +1,5 @@
 name: Feature or model request
-description: Propose a compute optimization, model adapter, or documentation improvement
+description: Propose a navigation computing optimization, model adapter, or documentation improvement
 title: "[Feature] "
 labels: ["enhancement"]
 body:

CAPABILITY.md

@@ -92,7 +92,7 @@ robot travels at most one increment past the cancel.
 
 ### `robonix/service/navigation/vln/telemetry`
 
-Read the compute-optimization measurements for a run. Empty `run_id` means the
+Read the navigation computing optimization measurements for a run. Empty `run_id` means the
 most recent run.
 
 Returns `{known, steps, sync_count, timeout_count, late_absorbed_count,
@@ -270,5 +270,5 @@ Call `navigate` once with the instruction and stop there. `navigate`,
 executor takes the returned `run_id`, polls the run to a terminal state and
 reports it; the caller does not have to sequence start → poll → cancel itself.
 
-`telemetry` is for reading the compute-optimization counters after a run, not
+`telemetry` is for reading the navigation computing optimization counters after a run, not
 for driving one.

CHANGELOG.md

@@ -2,6 +2,16 @@
 
 All notable public changes are recorded here.
 
+## 0.4.2 — 2026-08-03
+
+### Changed
+
+- public branding and method labels now say **Navigation Computing Optimization**
+  / **Navigation Computing** instead of Compute Optimization / Compute Skill.
+  Catalog tags and package keywords follow as `navigation-computing-optimization`.
+  Historical package ids, the pip distribution name, the `robonix_compute` import
+  tree, and `robonix-compute-*` CLI entry points are unchanged.
+
 ## 0.4.1 — 2026-07-29
 
 Fixes correctness defects exposed by running the trained InternVLA-N1 S1/S2
@@ -201,7 +211,7 @@ drop-in for a 0.2.0 deployment manifest.
 - documentation no longer calls this package a skill: the README title, the
   `CITATION.cff` / BibTeX title and the `robonix_compute` docstrings drop the
   `-Skill` suffix from the project name, and "What the Skill Optimizes" is now
-  "What the Runtime Optimizes". The benchmark method label `Compute Skill` is
+  "What the Runtime Optimizes". The benchmark method label `Navigation Computing` is
   deliberately unchanged — it is published data, appearing in
   `benchmarks/r2r_ce/results/*.csv`, `metadata.yaml`, `render_results.py` and the
   generated figures, and renaming it would desynchronize the tables from the

CITATION.cff

@@ -1,9 +1,9 @@
 cff-version: 1.2.0
 message: If you use this software, please cite the repository and the upstream InternNav/InternVLA work.
-title: RoboNix Compute Optimization
+title: RoboNix Navigation Computing Optimization
 type: software
-version: 0.4.1
-date-released: 2026-07-29
+version: 0.4.2
+date-released: 2026-08-03
 authors:
   - family-names: Cao
     given-names: Hangyu
@@ -17,5 +17,5 @@ keywords:
   - RoboNix
   - embodied AI
   - vision-language navigation
-  - compute optimization
+  - navigation computing optimization
   - edge inference

NOTICE

@@ -1,4 +1,4 @@
-RoboNix Compute Optimization Skill is an adaptive dual-system inference runtime
+RoboNix Navigation Computing Optimization is an adaptive dual-system inference runtime
 with an external RoboNix Skill boundary.
 
 The model adapters are designed to integrate with InternNav and InternVLA-N1

README.md

@@ -4,8 +4,8 @@
      Markdown between these tags as literal source text. -->
 <div align="center">
 <p><strong>This RoboNix Service is provided and maintained by Prof. Xiang Chen's group (<a href="https://if-lab-pku.github.io/">IFLab</a>), School of Computer Science, Peking University.</strong></p>
-<h1>RoboNix Compute Optimization</h1>
-<p><strong>An open-source compute optimization tool for RoboNix dual-system vision-language navigation</strong></p>
+<h1>RoboNix Navigation Computing Optimization</h1>
+<p><strong>An open-source navigation computing optimization toolkit for RoboNix dual-system VLN</strong></p>
 <p>
   <a href="README.zh-CN.md">简体中文</a> ·
   <a href="#what-this-adds">What this adds</a> ·
@@ -99,7 +99,7 @@ duplicated here.
   <img width="100%" src="docs/assets/benchmark_overview.svg" alt="R2R-CE benchmark overview">
 </div>
 
-Compute Skill optimizes the latency–accuracy trade-off rather than a single
+Navigation Computing optimizes the latency–accuracy trade-off rather than a single
 metric. On Orin+A100, SR remains on par with Edge Only (`62.8` versus `62.9`)
 while average step latency falls by 2.22×. Step Sync reaches a higher SR but
 blocks for `1644.5 ms` per step. The complete Orin and Thor tables are included
@@ -154,6 +154,9 @@ in [Benchmark Results](#benchmark-results).
 <a id="news"></a>
 ## 📰 News
 
+- **2026-08-03 — v0.4.2:** Public branding now uses **Navigation Computing
+  Optimization**; the benchmark method label is **Navigation Computing**. See
+  [CHANGELOG.md](CHANGELOG.md).
 - **2026-07-29 — v0.4.1:** Corrected the trained InternVLA-N1 service loop:
   one action per observed frame, S2 history replay across edge-only steps,
   action-only STOP / LOOK_DOWN handling, forced refresh after an exhausted
@@ -244,14 +247,14 @@ Full field list, episode selection criteria, and editing rules:
 
 Naive asynchronous execution is fast but may repeatedly consume stale semantic
 context. Full step synchronization keeps context fresh but blocks the control
-loop. Compute Skill switches between these behaviors according to runtime
+loop. Navigation Computing switches between these behaviors according to runtime
 state.
 
 <a id="architecture"></a>
 ## 🧠 Architecture
 
 <div align="center">
-  <img width="100%" src="docs/assets/compute_optimization_architecture.png" alt="Compute optimization runtime architecture">
+  <img width="100%" src="docs/assets/compute_optimization_architecture.png" alt="Navigation computing optimization runtime architecture">
 </div>
 
 1. **Fast local loop:** edge S1 consumes the latest observation and active
@@ -435,7 +438,7 @@ licensed assets were parsed and validated but no model result is claimed.
 | Real-model strict preflight | ✅ Executed | InternNav, Habitat, four model shards, S1 weights, depth checkpoint, data, ports, and two GPU IDs |
 | R2R one-episode dual-GPU run | ✅ Connectivity smoke | Completed navigation and produced result, progress, runtime, and telemetry artifacts; not benchmark evidence |
 | R2R-CE 1,839-episode tables | ✅ Project benchmark summary | Compact CSV and metadata are versioned; raw full-split logs are not bundled |
-| R2R difficulty / RxR / REVERIE proxy results | Data preparation only | No Compute Skill full-split result is claimed |
+| R2R difficulty / RxR / REVERIE proxy results | Data preparation only | No Navigation Computing full-split result is claimed |
 | Original REVERIE task | Not supported | Different MatterSim/object-grounding stack; not presented as REVERIE-CE support |
 
 Hosted CI covers the Python 3.9–3.11 test matrix, CPU quick start, CLI entry
@@ -747,7 +750,7 @@ The extended check expects:
 | RxR-CE English compatibility split | 3,669 | Data and nDTW ground truth validated |
 | REVERIE navigation-only proxy | 3,433 | Navigation proxy data validated |
 
-Passing the extended data check does not imply that Compute Skill has a
+Passing the extended data check does not imply that Navigation Computing has a
 full-split result on those optional datasets.
 
 <a id="step-4-strict-preflight"></a>
@@ -893,7 +896,7 @@ and orchestration remain external.
 | Reported edge devices | NVIDIA AGX Jetson Orin and Thor, `MAX_N` |
 | Navigation metrics | NE, SR, SPL |
 | System metrics | Average step latency, peak edge memory, peak cloud memory |
-| Compared strategies | Edge Only, Cloud Only, Naive ECC, Step Sync, Compute Skill |
+| Compared strategies | Edge Only, Cloud Only, Naive ECC, Step Sync, Navigation Computing |
 
 ### 1. Local one-episode connectivity run
 
@@ -989,7 +992,7 @@ python3 benchmarks/r2r_ce/render_results.py
 | Cloud Only | 4.05 | 64.3 | 58.5 | 128.4 ms | — | 16.62 GB |
 | Naive ECC | 4.76 | 56.7 | 45.1 | 202.0 ms | 0.60 GB | 16.03 GB |
 | Step Sync | 4.13 | 65.2 | 58.1 | 1644.5 ms | 0.60 GB | 16.03 GB |
-| **Compute Skill** | **4.18** | **62.8** | **57.8** | **224.4 ms** | **0.60 GB** | **16.03 GB** |
+| **Navigation Computing** | **4.18** | **62.8** | **57.8** | **224.4 ms** | **0.60 GB** | **16.03 GB** |
 
 ### Thor + A100
 
@@ -999,7 +1002,7 @@ python3 benchmarks/r2r_ce/render_results.py
 | Cloud Only | 4.05 | 64.3 | 58.5 | 128.4 ms | — | 16.62 GB |
 | Naive ECC | 4.72 | 56.7 | 45.4 | 164.6 ms | 0.60 GB | 16.03 GB |
 | Step Sync | 4.15 | 64.1 | 57.2 | 1328.0 ms | 0.60 GB | 16.03 GB |
-| **Compute Skill** | **4.14** | **63.1** | **58.1** | **166.3 ms** | **0.60 GB** | **16.03 GB** |
+| **Navigation Computing** | **4.14** | **63.1** | **58.1** | **166.3 ms** | **0.60 GB** | **16.03 GB** |
 
 ### Orin+A100 ablation
 
@@ -1019,7 +1022,7 @@ python3 benchmarks/r2r_ce/render_results.py
 
 Cloud Only is a compute upper bound and does not represent an edge-owned local
 control loop. Step Sync is a freshness upper bound with substantial blocking.
-Compute Skill targets the accuracy–latency trade-off.
+Navigation Computing targets the accuracy–latency trade-off.
 
 ### Reporting boundaries
 
@@ -1038,7 +1041,7 @@ Compute Skill targets the accuracy–latency trade-off.
 The optional extended profile is useful for future generalization studies, but
 the readiness levels are intentionally narrow:
 
-| Dataset | Data validation | Public Compute Skill result |
+| Dataset | Data validation | Public Navigation Computing result |
 | --- | --- | --- |
 | R2R-CE short / medium / long | Episode counts, ranges, and scene assets checked | Not claimed |
 | RxR-CE English compatibility split | 3,669 episodes and nDTW ground truth checked | Not claimed |
@@ -1231,9 +1234,9 @@ and citing it:
 ```bibtex
 @software{robonix_compute_optimization_2026,
   author  = {Cao, Hangyu and Zheng, Zihao},
-  title   = {RoboNix Compute Optimization},
+  title   = {RoboNix Navigation Computing Optimization},
   year    = {2026},
-  version = {0.4.1},
+  version = {0.4.2},
   url     = {https://github.com/i6bimua/service-navigation-vln-rbnx}
 }
 ```

README.zh-CN.md

@@ -3,8 +3,8 @@
      目录页上就会直接显示成源码。 -->
 <div align="center">
 <p><strong>本 RoboNix Service 由北京大学计算机学院陈翔教授课题组（<a href="https://if-lab-pku.github.io/">IFLab</a>）提供与维护。</strong></p>
-<h1>RoboNix Compute Optimization</h1>
-<p><strong>面向 RoboNix 双系统视觉语言导航的开源计算优化工具</strong></p>
+<h1>RoboNix Navigation Computing Optimization</h1>
+<p><strong>面向 RoboNix 双系统视觉语言导航的开源导航计算优化工具</strong></p>
 <p>
   <a href="README.md">English</a> ·
   <a href="#what-this-adds">为 RoboNix 带来了什么</a> ·
@@ -40,7 +40,7 @@ RoboNix 生命周期和机器人 I/O。
 运行时正确性、调度与 RoboNix 集成。
 
 目录中登记为 `robonix.service.navigation.vln`，因为它对 RoboNix 暴露的是一个长期运行
-的导航 Service；云边计算优化是该 Service 的核心实现。云端 S2 属于同一运行时，运行在
+的导航 Service；导航计算优化是该 Service 的核心实现。云端 S2 属于同一运行时，运行在
 机器人部署外的 GPU 主机上。HTTP 生命周期 API 仍为非 RoboNix 编排器保留。详见
 [RoboNix 集成边界](#robonix-integration-boundary)。
 
@@ -89,7 +89,7 @@ RoboNix TUI 的完整启动、指令输入、工具调用与终态演示放在
   <img width="100%" src="docs/assets/benchmark_overview.svg" alt="R2R-CE benchmark 结果总览">
 </div>
 
-Compute Skill 优化的是时延与精度的综合权衡，而不是单一指标。在 Orin+A100 上，SR 与 Edge Only 基本持平（`62.8` 对 `62.9`），平均单步时延降低 2.22×。Step Sync 的 SR 更高，但每步阻塞 `1644.5 ms`。Orin 与 Thor 完整表格见 [Benchmark 结果](#benchmark-results)。
+Navigation Computing 优化的是时延与精度的综合权衡，而不是单一指标。在 Orin+A100 上，SR 与 Edge Only 基本持平（`62.8` 对 `62.9`），平均单步时延降低 2.22×。Step Sync 的 SR 更高，但每步阻塞 `1644.5 ms`。Orin 与 Thor 完整表格见 [Benchmark 结果](#benchmark-results)。
 
 ### 按目标选择入口
 
@@ -110,7 +110,7 @@ Compute Skill 优化的是时延与精度的综合权衡，而不是单一指标
 - [项目动态](#news)
 - [系统效果](#results)
 - [Habitat 演示拍摄](#demo-filming)
-- [核心计算优化](#what-the-runtime-optimizes)
+- [核心导航计算优化](#what-the-runtime-optimizes)
 - [系统架构](#architecture)
 - [运行图片](#running-images)
 - [RoboNix 集成边界](#robonix-integration-boundary)
@@ -140,6 +140,9 @@ Compute Skill 优化的是时延与精度的综合权衡，而不是单一指标
 <a id="news"></a>
 ## 📰 项目动态
 
+- **2026-08-03 — v0.4.2：**对外品牌统一为 **Navigation Computing Optimization**
+  （导航计算优化）；benchmark 方法名改为 **Navigation Computing**。详见
+  [CHANGELOG.md](CHANGELOG.md)。
 - **2026-07-29 — v0.4.1：**修正训练后 InternVLA-N1 接入完整 Service 闭环时暴露的
   问题：每个观测只执行一个动作、边端独立步骤回放进 S2 历史、正确处理纯动作
   STOP / LOOK_DOWN、动作块耗尽后强制刷新、RGB/深度输入转换，以及可复现的 Habitat
@@ -190,7 +193,7 @@ python scripts/demo/make_demo_reels.py --mode speed \
 字段、选片标准与剪辑规则见 [benchmarks/r2r_ce/DEMO_FILMING.md](benchmarks/r2r_ce/DEMO_FILMING.md)。
 
 <a id="what-the-runtime-optimizes"></a>
-## 🧩 核心计算优化
+## 🧩 核心导航计算优化
 
 | 机制 | 运行时作用 |
 | --- | --- |
@@ -200,13 +203,13 @@ python scripts/demo/make_demo_reels.py --mode speed \
 | **自适应缺失处理** | 超时预测与 latent 复用限制网络阻塞，同时保留迟到响应。 |
 | **完整遥测** | 逐步记录 S1/S2 时延、同步、超时、复用、载荷大小与导航指标。 |
 
-朴素异步执行速度快，但可能持续消费过时语义上下文；全步同步保持上下文新鲜，却会阻塞控制闭环。Compute Skill 根据运行时状态在两种行为之间切换。
+朴素异步执行速度快，但可能持续消费过时语义上下文；全步同步保持上下文新鲜，却会阻塞控制闭环。Navigation Computing 根据运行时状态在两种行为之间切换。
 
 <a id="architecture"></a>
 ## 🧠 系统架构
 
 <div align="center">
-  <img width="100%" src="docs/assets/compute_optimization_architecture.png" alt="计算优化运行时架构">
+  <img width="100%" src="docs/assets/compute_optimization_architecture.png" alt="导航计算优化运行时架构">
 </div>
 
 1. **端侧快速闭环：**S1 融合最新观测和 active context 生成动作，不等待云端推理。
@@ -367,7 +370,7 @@ conda create -n rbnx-ros -c robostack-staging -c conda-forge python=3.11 \
 | 真实模型严格预检 | ✅ 已执行 | InternNav、Habitat、4 个模型分片、S1、深度权重、数据、端口和双 GPU ID |
 | R2R 单 episode 双 GPU 运行 | ✅ 连通性 smoke | 完成导航并产出 result、progress、runtime 与 telemetry；不作为 benchmark 证据 |
 | R2R-CE 1,839 episode 表格 | ✅ 项目 benchmark 摘要 | 版本化保存紧凑 CSV 与 metadata；不打包完整原始日志 |
-| R2R 难度子集 / RxR / REVERIE proxy 结果 | 仅数据准备 | 不宣称 Compute Skill 已有这些数据集的 full-split 结果 |
+| R2R 难度子集 / RxR / REVERIE proxy 结果 | 仅数据准备 | 不宣称 Navigation Computing 已有这些数据集的 full-split 结果 |
 | 原版 REVERIE | 不支持 | 使用不同的 MatterSim/object-grounding 技术栈，不作为 REVERIE-CE 支持项 |
 
 托管 CI 覆盖 Python 3.9–3.11 测试矩阵、CPU Quick Start、全部 CLI、文档链接、SVG XML、结果图确定性生成、包构建和发布审计。GPU/Habitat 检查使用显式的本地或自托管流程。
@@ -659,7 +662,7 @@ robonix-compute-data-check \
 | RxR-CE English compatibility split | 3,669 | 数据与 nDTW ground truth 已验证 |
 | REVERIE navigation-only proxy | 3,433 | 导航代理数据已验证 |
 
-扩展数据检查通过，不代表 Compute Skill 已具备这些可选数据集的完整结果。
+扩展数据检查通过，不代表 Navigation Computing 已具备这些可选数据集的完整结果。
 
 <a id="step-4-strict-preflight"></a>
 ## 🔍 步骤 4：严格预检
@@ -799,7 +802,7 @@ curl -X POST http://127.0.0.1:8090/step \
 | 报告的端侧设备 | NVIDIA AGX Jetson Orin、Thor，`MAX_N` |
 | 导航指标 | NE、SR、SPL |
 | 系统指标 | 平均单步时延、端侧峰值内存、云端峰值内存 |
-| 对比策略 | Edge Only、Cloud Only、Naive ECC、Step Sync、Compute Skill |
+| 对比策略 | Edge Only、Cloud Only、Naive ECC、Step Sync、Navigation Computing |
 
 ### 1. 本地单 episode 连通性运行
 
@@ -888,7 +891,7 @@ python3 benchmarks/r2r_ce/render_results.py
 | Cloud Only | 4.05 | 64.3 | 58.5 | 128.4 ms | — | 16.62 GB |
 | Naive ECC | 4.76 | 56.7 | 45.1 | 202.0 ms | 0.60 GB | 16.03 GB |
 | Step Sync | 4.13 | 65.2 | 58.1 | 1644.5 ms | 0.60 GB | 16.03 GB |
-| **Compute Skill** | **4.18** | **62.8** | **57.8** | **224.4 ms** | **0.60 GB** | **16.03 GB** |
+| **Navigation Computing** | **4.18** | **62.8** | **57.8** | **224.4 ms** | **0.60 GB** | **16.03 GB** |
 
 ### Thor + A100
 
@@ -898,7 +901,7 @@ python3 benchmarks/r2r_ce/render_results.py
 | Cloud Only | 4.05 | 64.3 | 58.5 | 128.4 ms | — | 16.62 GB |
 | Naive ECC | 4.72 | 56.7 | 45.4 | 164.6 ms | 0.60 GB | 16.03 GB |
 | Step Sync | 4.15 | 64.1 | 57.2 | 1328.0 ms | 0.60 GB | 16.03 GB |
-| **Compute Skill** | **4.14** | **63.1** | **58.1** | **166.3 ms** | **0.60 GB** | **16.03 GB** |
+| **Navigation Computing** | **4.14** | **63.1** | **58.1** | **166.3 ms** | **0.60 GB** | **16.03 GB** |
 
 ### Orin+A100 消融
 
@@ -916,7 +919,7 @@ python3 benchmarks/r2r_ce/render_results.py
 | 控制状态与 counters | 端侧内存 `< 1 KB` |
 | 切换与编排逻辑 | 相比模型推理可忽略 |
 
-Cloud Only 是计算上界，不代表端侧本地闭环。Step Sync 是上下文新鲜度上界，但阻塞开销较大。Compute Skill 关注精度—时延权衡。
+Cloud Only 是计算上界，不代表端侧本地闭环。Step Sync 是上下文新鲜度上界，但阻塞开销较大。Navigation Computing 关注精度—时延权衡。
 
 ### 报告边界
 
@@ -933,7 +936,7 @@ Cloud Only 是计算上界，不代表端侧本地闭环。Step Sync 是上下
 
 可选扩展 profile 可用于后续泛化分析，但其就绪范围被严格限制：
 
-| 数据集 | 数据验证 | 公开 Compute Skill 结果 |
+| 数据集 | 数据验证 | 公开 Navigation Computing 结果 |
 | --- | --- | --- |
 | R2R-CE short / medium / long | Episode 数、范围与场景资产已检查 | 不宣称 |
 | RxR-CE English compatibility split | 3,669 episodes 与 nDTW ground truth 已检查 | 不宣称 |
@@ -1111,9 +1114,9 @@ python3 -m build
 ```bibtex
 @software{robonix_compute_optimization_2026,
   author  = {Cao, Hangyu and Zheng, Zihao},
-  title   = {RoboNix Compute Optimization},
+  title   = {RoboNix Navigation Computing Optimization},
   year    = {2026},
-  version = {0.4.1},
+  version = {0.4.2},
   url     = {https://github.com/i6bimua/service-navigation-vln-rbnx}
 }
 ```

THIRD_PARTY_LICENSES.md

@@ -1,6 +1,6 @@
 # Third-Party Components
 
-RoboNix Compute Optimization keeps heavyweight third-party projects and checkpoints outside this
+RoboNix Navigation Computing Optimization keeps heavyweight third-party projects and checkpoints outside this
 repository. Users should install and cite them according to their own licenses.
 
 | Component | Use | Distribution policy |

benchmarks/r2r_ce/README.md

@@ -11,7 +11,7 @@ full-split logs.
 - Backbone: InternVLA-N1 DualVLN with original weights
 - Cloud: NVIDIA A100
 - Edge: NVIDIA AGX Jetson Orin or Thor in `MAX_N`
-- Strategies: Edge Only, Cloud Only, Naive ECC, Step Sync, and Compute Skill
+- Strategies: Edge Only, Cloud Only, Naive ECC, Step Sync, and Navigation Computing
 
 The CSV files are the structured source for the project benchmark summary. See
 `metadata.yaml` for the complete evaluation conditions.