start-jsk / jsk_apc

Other
36 stars 35 forks source link

TODOs for ARC2017 #2035

Closed pazeshun closed 7 years ago

pazeshun commented 7 years ago

Critical

State TODO Detail Creator Assignees
done 既知物体(40個+背景)の物体セグメンテーション #2028 @wkentaro @wkentaro
ongoing 新しい棚の作成 #2017 @knorth55 @knorth55
ongoing 新グリッパ(Gripper-V5)の作成 #2036 @pazeshun @pazeshun
na Gripper-V5での把持計画(pick) #2038 @pazeshun @pazeshun
na Gripper-V5での把持計画(stow) #2043 @knorth55 @knorth55
done pickタスク実現 #2053 @wkentaro @knorth55
done stowタスク実現 #2053 @wkentaro @knorth55

Not Critical

TODO Detail Creator Assignees
透明な棚など認識や動作に有効な棚の作成 https://github.com/start-jsk/jsk_apc/issues/2017#issuecomment-287961884 @k-okada
近接センサを搭載したグリッパによる手探り動作 #2047 @knorth55
煩雑環境での既知物体セグメンテーション #2028 @wkentaro
短時間(~45分)で新規物体を認識可能にする手法の開発 #2037 @wkentaro

2017-03-25T18:30 Updated by @wkentaro 2017-03-25T20:28 Updated by @pazeshun 2017-03-26T11:59 Updated by @knorth55 2017-03-26T13:36 Updated by @knorth55 2017-03-27T03:11 Updated by @wkentaro 2017-04-04T23:40 Updated by @knorth55 2017-04-11T15:30 Updated by @wkentaro 2017-05-10T18:21 Updated by @knorth55 2017-06-13T18:27 Updated by @knorth55

pazeshun commented 7 years ago

各位、適宜加筆修正をお願いします

pazeshun commented 7 years ago

グリッパの課題の詳細を #2036 に移動しました

k-okada commented 7 years ago

近接センサを搭載したグリッパによる手探り動作 これ、もし、うまくいくなら、これ、すごく重要.これが出来るようになりましょう.

knorth55 commented 7 years ago

近接センサを搭載したグリッパによる手探り動作

これ興味があります。 把持計画と一緒に少し考えたいです。

wkentaro commented 7 years ago

ここに書きましたが、https://github.com/start-jsk/jsk_apc/issues/2028#issuecomment-289302473 Bin内のコンテクストありなら既知物体をセグメントできるようになり、今のでもう使える状態ではあると思います。 より煩雑な環境で認識することが必要な場合には、より多くのデータセット作成と再学習をする必要があるので、言ってください。

k-okada commented 7 years ago

近接センサ:

を書いてみましょう.if文でよいです.これで、https://youtu.be/-dRwReGMeb4?t=104 ができるはずです. 同じ方法は、このまえのmbzircでも有効性が実証されています.https://youtu.be/EsqRXrFcaS4?t=12501

一回プログラムが動いたら綺麗にプログラムをまとめることが出来るはずで、 https://www.jstage.jst.go.jp/article/biomechanisms/1/0/1_KJ00004274842/_pdf に考え方が有ります. http://pwp.gatech.edu/wp-content/uploads/sites/231/2016/05/auro10_ele.pdf, Fig 5 も同様. 最近だとcram https://www.open-ease.org/papers/beetz10cram.pdf もこれがある. @furushchev

furushchev commented 7 years ago

@k-okada 他のチームが制限時間いっぱいでも回すことができない中颯爽と現れたこのチームは衝撃でしたね。 cramでは基本的に先生が上で書かれたようなif文がいっぱいある目標状態ベースで書かれていると思います。 動作の段階に応じて使えるセンサや認識の仕方も変わってくると思うので、目標状態をうまく分離して遷移を設計する必要があるように思います。(APCだと比較的単純に設計できそう) ドイツにいた時に起きたデモでの失敗あるあるだと、目標状態の監視と目標到達動作を並列に実行している時に各状態の積集合がゼロになってあちらを立てればこちらが立たずという感じになってリスタートということが割と頻繁に起きていた印象がありました。 というのがコメントです。 (実は現在某Robocupでも上手くこの辺を統合できないかと考えています。)

k-okada commented 7 years ago

https://github.com/cram-code/cram_tutorials/blob/ce9c2281891ffaaf0bf1b8ef09e32635f8142720/cram_beginner_tutorial/src/simple-plans.lisp#L38-L46 とかちがうかな.

-- ◉ Kei Okada

2017年3月28日 11:18 Furushchev notifications@github.com:

@k-okada https://github.com/k-okada 他のチームが制限時間いっぱいでも回すことができない中颯爽と現れたこのチームは衝撃でしたね。 cramでは基本的に先生が上で書かれたようなif文がいっぱいある目標状態ベースで書かれていると思います。 動作の段階に応じて使えるセンサや認識の仕方も変わってくると思うので、目標状態をうまく分離して遷移を設計する必要があるように思いま す。(APCだと比較的単純に設計できそう) ドイツにいた時に起きたデモでの失敗あるあるだと、目標状態の監視と目標到達動作を並列に実行している時に各状態の 積集合がゼロになってあちらを立てればこちらが立たずという感じになってリスタートということが割と頻繁に起きていた印象がありました。 というのがコメントです。 (実は現在某Robocupでも上手くこの辺を統合できないかと考えています。)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/start-jsk/jsk_apc/issues/2035#issuecomment-289642388, or mute the thread https://github.com/notifications/unsubscribe-auth/AAeG3A8JwQ8r9SL86a5ZwEftUMs3IJQQks5rqG3wgaJpZM4MpAdM .

furushchev commented 7 years ago

@k-okada それです。僕の帰国報告スライドにも載せてありましたが、 http://cram-system.org/tutorials/beginner/simple_plans に明解なドキュメントとチュートリアルがあります。

k-okada commented 7 years ago

修士の人はif文まで、さっそうとタスクを片付ける事が期待されますね. 博士の人は実戦投入は置いておいて、guradマクロでこうなります、と論文にかけるところまで、うまくまとめることが期待されますね.

-- ◉ Kei Okada

2017-03-28 22:19 GMT+09:00 Furushchev notifications@github.com:

@k-okada https://github.com/k-okada それです。僕の帰国報告スライドにも載せてありましたが、 http://cram-system.org/tutorials/beginner/simple_plans に明解なドキュメントとチュートリアルがあります。

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/start-jsk/jsk_apc/issues/2035#issuecomment-289766721, or mute the thread https://github.com/notifications/unsubscribe-auth/AAeG3B_JhySwwwEF1dcZPyUAnNqJNrSfks5rqQjKgaJpZM4MpAdM .

knorth55 commented 7 years ago

@k-okada mbzircの動画を見ましたが、井上先生の書かれていることと同じようなことを行っていますね 一度だけみたあとつかむ方式(ルックアンドムーブ?)だと、認識のあとはこの手法が有効そうですね。 見ている場合は

knorth55 commented 7 years ago

@k-okada https://github.com/RoboticMaterials/FA-I-sensor/pull/6 これでFA-I sensor baxter gripperの値がrosのtopicでとれるようになりました

k-okada commented 7 years ago

ナイス 物体の表面を手が通り過ぎた時の結果を撮って見ましょう

2017年4月4日(火) 23:12 Shingo Kitagawa notifications@github.com:

@k-okada https://github.com/k-okada RoboticMaterials/FA-I-sensor#6 https://github.com/RoboticMaterials/FA-I-sensor/pull/6 これでFA-I sensor baxter gripperの値がrosのtopicでとれるようになりました

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/start-jsk/jsk_apc/issues/2035#issuecomment-291512226, or mute the thread https://github.com/notifications/unsubscribe-auth/AAeG3GoQP3dgOy7n3UkEqCBbfpOJ_Yd1ks5rsk-pgaJpZM4MpAdM .

--

◉ Kei Okada

wkentaro commented 7 years ago

@knorth55 ここじゃなくてIssue立てて下さいm( )m

k-okada commented 7 years ago

ちなみにrviz にビジュアライズすると言う優先度最低のタスクもありますね Eusでモデル作って色つけてマーカーパブリッシュかな m1で2日、b4で1週間ぐらいでしょうか

2017年4月4日(火) 23:28 Kei Okada k-okada@jsk.t.u-tokyo.ac.jp:

ナイス 物体の表面を手が通り過ぎた時の結果を撮って見ましょう

2017年4月4日(火) 23:12 Shingo Kitagawa notifications@github.com:

@k-okada https://github.com/k-okada RoboticMaterials/FA-I-sensor#6 https://github.com/RoboticMaterials/FA-I-sensor/pull/6 これでFA-I sensor baxter gripperの値がrosのtopicでとれるようになりました

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/start-jsk/jsk_apc/issues/2035#issuecomment-291512226, or mute the thread https://github.com/notifications/unsubscribe-auth/AAeG3GoQP3dgOy7n3UkEqCBbfpOJ_Yd1ks5rsk-pgaJpZM4MpAdM .

--

◉ Kei Okada

--

◉ Kei Okada

wkentaro commented 7 years ago

何のビジュアライズでしょうか。

k-okada commented 7 years ago

これでFA-I sensor baxter gripperの値がrosのtopicでとれるようになりました

です 2017年4月4日(火) 23:32 Kentaro Wada notifications@github.com:

何のビジュアライズでしょうか。

You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/start-jsk/jsk_apc/issues/2035#issuecomment-291518666, or mute the thread https://github.com/notifications/unsubscribe-auth/AAeG3KThvi4WKNU089LbH3Dm64MAidSEks5rslSNgaJpZM4MpAdM .

--

◉ Kei Okada

knorth55 commented 7 years ago

@pazeshun この前話し合った内容を更新してもらえますか?

pazeshun commented 7 years ago

すみません、この前というと、人工知能学会前の話でしょうか?

wkentaro commented 7 years ago

We have finished all critical tasks at 3.2. Now we move forward to improvement of the hardware, calibration and recognition algorithms.