Practise teleoperation in simulation and keep an exact record.シミュレーションで遠隔操作を練習し、正確な記録を残す。
Start in simulation, not on expensive hardware. Pick one robot, one task and one input device; practise; record each session as structured evidence; review the failures.
This curriculum is original BSV training material. It links NVIDIA's official setup and does not redistribute NVIDIA code. BSV has not run this curriculum itself yet.
Not a government licence or a manufacturer certification.国の免許やメーカーの認定ではありません。
Not permission to operate real hardware. Real-hardware control needs the robot owner's authorisation and their local safety rules (supervisor and emergency stop where required).実機を操作してよいという許可ではありません。実機の操作には、ロボットの所有者の許可と、その場の安全ルール(必要に応じて監督者・非常停止)が必要です。
Not proof of general ability across robots. A record covers one exact task, robot, runtime, input device and revision.ロボット全般の能力の証明ではありません。記録の対象は、特定のタスク・ロボット・実行環境・入力装置・リビジョンだけです。
Self-reported. A BSV review only means the evidence was looked at; it never upgrades the record.自己申告です。BSVの確認は証跡を見たという意味だけで、記録の格が上がることはありません。
Isaac Teleop SO-101 Simulation Pilot Starter
Runtime: NVIDIA Isaac Lab + Isaac Teleop · Embodiment: SO-101 simulated follower · Input: XR controller or SO-101 leader, depending on task · Environment: SIMULATION · curriculum JSON · schema
Before you start始める前に
Install Isaac Lab using NVIDIA's current official instructions
Install/build the Isaac Teleop components required by the chosen input device
Use the current official task id matching the input device
Run only in simulation for this starter curriculum
Launch the correct simulation task and confirm input, visualization and reset behavior before recording.
Evidence
runtime version
task id
input device
successful reset
Pass criteria (tick what you met)
Basic control
Demonstrate bounded end-effector/gripper control without recording a scored episode.
Evidence
practice note
known control limitations
Pass criteria (tick what you met)
Record first ten demonstrations
Record ten task demonstrations using the official demonstration-recording path.
Evidence
dataset reference
attempted count
success count
failure count
Pass criteria (tick what you met)
Failure and recovery review
Review failed episodes and identify repeatable failure categories instead of deleting them.
Evidence
failure taxonomy
at least one reviewed failed episode if failures occurred
Pass criteria (tick what you met)
Repeatability block
Run a second session and compare completion, failure and reset consistency.
Evidence
second session evidence id
comparison note
Pass criteria (tick what you met)
Completion status in this curriculum: SELF_REPORTED (minimum reviewed sessions: 0).このカリキュラムの修了状態: 自己申告(必要な確認済みセッション数: 0)。
Practice record (session evidence)練習記録(セッションの証跡)
Fill this in after a simulation session. It builds two JSON files: session evidence (teleop-session-evidence v0.1, review status UNREVIEWED) and a practice record (robot-pilot-profile v0.1, SELF_REPORTED). It stays in this browser unless you press Send.シミュレーションのセッション後に入力します。2つのJSONを作ります: セッションの証跡(teleop-session-evidence v0.1、確認状態 UNREVIEWED)と練習記録(robot-pilot-profile v0.1、SELF_REPORTED)。「送る」を押さない限り、このブラウザーの中だけにあります。
Check a saved file保存したファイルを確認
Pick a session-evidence or practice-record JSON you downloaded here. It is checked against the published schema in this browser; nothing is uploaded. A file cannot raise its own review status.ここでダウンロードした証跡または練習記録のJSONを選んでください。公開しているスキーマに合っているかを、このブラウザの中だけで確認します。アップロードはしません。ファイルを書き換えても確認の状態は上がりません。
Totals over the sessions you saved with “Save in this browser” (up to 20). Computed here from your own entries; nothing is uploaded, estimated or reviewed. Sessions whose counts do not add up are flagged, not corrected.「このブラウザーに保存」で保存したセッション(最大20件)の合計です。入力した値からこの場で計算します。アップロード・推計・確認はしていません。件数が合わないセッションは直さずに表示します。
Taskタスク
Sessionsセッション
Attempted試行
Successful成功
Failed失敗
Recoveryリカバリー
Success rate成功率
Only files valid against the published session-evidence schema are added; duplicates (same evidenceId) are skipped; the log keeps the last 20. Checked and stored in this browser only; nothing uploaded.公開スキーマに合うセッション記録だけを追加します。同じ evidenceId は追加しません。ログは最新20件まで。確認も保存もこのブラウザ内だけで、アップロードはしません。
The CSV is built in this browser from the saved files only: one row per session, values as saved, nothing uploaded.CSVは保存したファイルだけから、このブラウザ内で作ります。1行が1セッションで、値は保存したとおりです。何も送信しません。
Teleop recipe libraryテレオペ手順ライブラリ
Reusable operating procedures for practice sessions: SOP, reset, failure taxonomy, acceptance criteria, export map and replay checklist. Original BSV material, simulation only, not run by BSV. Item text is in English.練習セッションで使い回せる運用手順です(標準手順・リセット・失敗の分類・採用条件・出力の対応表・リプレイ確認)。BSVオリジナルの資料で、シミュレーション専用です。BSVはまだ実行していません。各項目の本文は英語です。
Written by BSV from the curriculum below; BSV has not run this procedure in Isaac Lab or Isaac Teleop. Follow NVIDIA's current official instructions wherever they differ.
Standard procedure (before / during / after)標準手順(前・中・後)
Before開始前
Note the Isaac Lab and Isaac Teleop versions (or source revisions) and the official task id you will run.
Launch the task, confirm the input device moves the simulated follower, and run one reset before recording.
Decide the session goal (attempted episodes, time box) and write it in the session note.
Duringセッション中
Record every attempted episode, including failures; do not delete failed episodes.
Label each episode right after it ends: SUCCESS, FAILURE or PARTIAL, and a failure category when it is not a success.
Reset between episodes using the documented reset path; note any reset that did not return the scene to its start state.
Stop the session if the simulator becomes unstable, and note it as a safety/reset event.
After終了後
Count attempted, successful and failed episodes and episodes that needed recovery.
Review at least one failed episode and write one concrete change for the next session.
Build the session evidence and practice record on the Academy page; keep the dataset reference you recorded.
Reset and recoveryリセットと復旧
Disengage control before resetting.
Use the task's documented reset, not a manual drag of objects.
Confirm the follower and objects are back at the start state before engaging control again.
If a reset fails twice, end the episode as FAILURE with category reset-failure and note it.
If control behaves unexpectedly, disengage first, then reset.
After a tracking-loss or input-lag failure, check the device before the next episode and note what you changed.
Count an episode as a recovery episode when you had to recover the object or scene mid-attempt.
Failure taxonomy and labels失敗の分類とラベル
SUCCESS · FAILURE · PARTIAL
id
Meaning意味
grasp-miss
Gripper closed without holding the object
drop
Object was held, then dropped before the goal
collision
Follower or object hit the scene in a way that ended the attempt
goal-miss
Object reached the area but not the goal pose or region
timeout
Attempt ran past the time box you set
tracking-loss
Input device tracking was lost or jumped
input-lag
Control lag made the attempt uncontrollable
reset-failure
Scene did not return to its start state
sim-instability
Simulator physics or rendering became unstable
operator-abort
Operator stopped the attempt on purpose
Episode acceptance and replay checklist採用条件とリプレイ確認
The episode was recorded through the official demonstration-recording path.
It has a label (SUCCESS, FAILURE or PARTIAL) and, unless SUCCESS, a failure category from this taxonomy.
It started from a confirmed reset state.
It is kept in the dataset even when it failed; exclusions are written in the session note with a reason.
Play back at least one SUCCESS and one FAILURE episode if you have both.
Check that the label matches what happened in the replay.
Check that the failure category is the first thing that went wrong, not the last.
Note any episode whose replay does not match the recorded label, and fix the label before review.
Controller mapping and retargeting settings: they depend on your input device and the official task id, so record the ones you used instead of copying a preset.
Real-hardware steps: this recipe is simulation only. Real-hardware authority belongs to the hardware owner or partner.
Tooling opportunities for engineersエンジニア向け: ツールの空き
What the BSV teleop material covers today, per tooling kind, and what it leaves out. Counted from the recipe and curriculum files on this site. This is not a demand figure: BSV has no count of people asking for these.ツールの種類ごとに、BSVの遠隔操作の資料で今カバーしていることと、していないことです。このサイトのレシピとカリキュラムのファイルから数えています。需要の数字ではありません(求めている人数をBSVは数えていません)。
Kind種類
BSV todayBSVの現状
Controller mappingコントローラーの割り当て
Not included. Recipe text: “Controller mapping and retargeting settings: they depend on your input device and the official task id, so record the ones you used instead of copying a preset.” The evidence file only records the device used (field inputDevice).含みません(レシピに「入力機器と公式タスクIDに依存するため、プリセットを写さず使った設定を記録する」とあります)。証跡ファイルには使った機器(inputDevice)だけを記録します。
Retargeting configsリターゲティング設定
Not included (same recipe line as above). No BSV retargeting preset exists.含みません(上と同じ記載)。BSVのリターゲティング用プリセットはありません。
Dashboardsダッシュボード
Practice log on this page (#log): totals from evidence files saved in this browser only, downloadable as CSV. No shared or hosted dashboard.このページの練習ログ(#log): このブラウザに保存した証跡ファイルからの合計だけです(CSVで保存できます)。共有・ホスト型のダッシュボードはありません。
Annotation and QA toolingアノテーション・QAのツール
Written lists only: 10 failure ids, 3 annotation labels, 4 episode-acceptance criteria; “Check a saved file” (#check) checks a JSON file against the schema. No labelling tool.文書のリストだけです: 失敗ID 10件、アノテーションラベル 3件、採用条件 4件。「ファイルを確認」(#check)でJSONをスキーマと照合できます。ラベル付けツールはありません。
Replay toolsリプレイツール
A replay checklist (4 items). No replay tool.リプレイの確認リスト(4項目)だけです。リプレイツールはありません。
Simulation scenesシミュレーションのシーン
None from BSV. The curriculum points at 2 official SO-101 task ids published by NVIDIA.BSVのシーンはありません。カリキュラムはNVIDIAが公開しているSO-101の公式タスクID 2件を案内します。
Data conversionデータ変換
An export map (9 fields) from a session to teleop-session-evidence v0.1. No converter to or from other dataset formats.セッションからteleop-session-evidence v0.1への対応表(9項目)だけです。他のデータ形式との変換ツールはありません。
Approved public requests from the Request Market in the robot-pilot area, read live from the request store. Nothing is seeded or estimated. A mission is a request: no escrow, no contract and no permission to operate real hardware.Request Marketのロボットパイロット分野で承認・公開されたリクエストを、リクエストの保存先からそのまま読み込みます。見本や推計は入れていません。ミッションはリクエストであり、エスクロー・契約・実機を操作する許可ではありません。
Need a pilot or a dataset? Request a missionパイロットやデータが必要なら: ミッションをリクエスト
Missions go through the Request Market (request only, no escrow). Describe the task, robot, teleop interface, environment, data and episode target. Real-hardware missions stay under the robot owner's authorisation and safety rules.ミッションはRequest Market(リクエストのみ・エスクローなし)から送ります。タスク・ロボット・遠隔操作の方法・環境・必要なデータ・エピソード数を書いてください。実機のミッションは、ロボットの所有者の許可と安全ルールのもとで行います。