在測試機器人行走軌跡時，我學到的三件事
第一件事：安全≠安心，標準與體感存在落差
原本針對靜態障礙物的 ISO 測試，核心指標是確保機器人在偵測到物體時能絕對煞停。這在數據驗證上無懈可擊，但在實際場域中，這種「直直走、急煞車」的行為模式，會讓周圍的人類感到焦慮。
符合規矩的機器人若動作太過僵硬或機械化，反而讓人難以預測它的下一步，進而降低了人類對其「智慧程度」的評價。
我真正學到的是：「通過標準」只能證明不違規；「被接受」才代表可共處。安全是底線，安心是門檻。
第二件事：說服力來自「預判性」與 Jerk 的平滑
在這次調整中，我們發現關鍵差異在於「預判性」與「加加速度（Jerk）」的平滑處理。真正的說服力，來自於移動意圖的提早展現。
若機器人能在接觸障礙物前三公尺就開始微幅調整航向，而非等到最後一刻才大幅度轉彎，使用者就會直觀地認為它「看懂了路況」。
這種細微的軌跡差異，無法單純用避障成功率（Success Rate）來衡量，但它決定了產品是否能融入人類生活圈。
只追求通過
- 直線前進、末端大轉彎
- 急煞、停停走走
- 人覺得「不可預測」
追求說服力
- 提前微調、連續小修正
- Jerk 平滑、速度變化自然
- 人覺得「懂環境、很協調」
第三件事：把人類心理學翻譯成程式碼，才會被信任
工程師看的是座標與誤差值，用戶看的是意圖與協調性。這次測試證明，優化演算法不僅是解決數學問題，更多時候是將人類的行為心理學，翻譯成機器讀得懂的程式碼。
一句話總結：當機器人的軌跡能「提前表態」，人就更願意把它當成隊友，而不是一台會突然動作的設備。
可落地的測試觀察清單
- ✦急煞是否過度頻繁：通過測試不代表可共處，需觀察旁人壓力與退讓行為。
- ✦提前修正距離是否一致：例如在障礙物前 2–3 公尺開始微調，意圖更可被理解。
- ✦速度曲線是否平滑：特別檢視 Jerk，避免「突然加速／突然停」造成不適與不信任。
- ✦轉彎幅度是否合理：少做末端大轉彎，多做連續小修正，整體更自然。
- ✦指標不只 Success Rate：加入「可預測性」「協調性」等體感維度（可用問卷/行為觀察量化）。
3 Things I Learned When Testing Robot Trajectories
1. Safety ≠ Peace of Mind: The Gap Between Standards and Experience
In original ISO tests for static obstacles, the core metric is ensuring the robot can absolutely stop when an object is detected. This is flawless in data validation, but in real-world scenarios, this "walk straight, brake suddenly" behavioral pattern makes surrounding humans feel anxious.
If a compliant robot's movements are too rigid or mechanical, it becomes hard for humans to predict its next step, which in turn lowers their evaluation of its "intelligence level."
What I truly learned: "Passing the standard" only proves there's no violation; "being accepted" means coexistence is possible. Safety is the baseline; peace of mind is the threshold.
2. Persuasiveness Comes from "Anticipation" and Smooth Jerk
During our adjustments, we found that the key difference lies in "anticipation" and the smooth processing of "Jerk" (rate of change of acceleration). True persuasiveness comes from the early display of movement intent.
If a robot can begin to slightly adjust its course three meters before encountering an obstacle, rather than making a sharp turn at the last second, users will intuitively feel that it "understands the road conditions."
This subtle difference in trajectory cannot be measured purely by Success Rate, but it determines whether the product can integrate into the human living sphere.
Pursuing Only Passage
- Moves straight, sharp turns at the end
- Sudden braking, stop-and-go
- Humans perceive it as "unpredictable"
Pursuing Persuasiveness
- Early tweaks, continuous minor corrections
- Smooth Jerk, natural speed changes
- Humans perceive it as "aware and coordinated"
3. Translating Human Psychology into Code to Build Trust
Engineers look at coordinates and error values; users look at intent and coordination. This test proved that optimizing algorithms is not just about solving math problems; more often, it is about translating human behavioral psychology into code that machines can read.
One-sentence summary: When a robot's trajectory can "express its intent early," humans are much more willing to treat it as a teammate rather than a device that moves abruptly.
Actionable Testing Observation Checklist
- ✦Is sudden braking too frequent? Passing a test doesn't equal coexistability. Observe the stress and yielding behavior of bystanders.
- ✦Is the early correction distance consistent? E.g., starting to tweak the path 2–3 meters before an obstacle makes the intent much easier to understand.
- ✦Is the speed curve smooth? Specifically check the Jerk metric to avoid "sudden acceleration/stopping" which causes discomfort and distrust.
- ✦Are the turn radiuses reasonable? Fewer large, sharp turns at the end, and more continuous minor corrections make the overall movement more natural.
- ✦Metrics beyond Success Rate: Introduce experiential dimensions like "Predictability" and "Coordination" (can be quantified via surveys or behavioral observation).