🖐️ 미시간대, 센서별 촉각 시뮬레이션 자체를 생략
SBLR은 시뮬레이터 안의 "오라클" 센서로 학습한 뒤 실제 촉각 데이터를 추론 시점에 정렬합니다. 제로샷 성공률 85~97.5%, 페그 삽입·기어 조립 과제에서 검증. (11/8)
https://www.alphaxiv.org/abs/2611.tactile-sim2real-latent-reconstruction
SBLR은 시뮬레이터 안의 "오라클" 센서로 학습한 뒤 실제 촉각 데이터를 추론 시점에 정렬합니다. 제로샷 성공률 85~97.5%, 페그 삽입·기어 조립 과제에서 검증. (11/8)
https://www.alphaxiv.org/abs/2611.tactile-sim2real-latent-reconstruction
November 9, 2026 at 1:04 AM
🖐️ 미시간대, 센서별 촉각 시뮬레이션 자체를 생략
SBLR은 시뮬레이터 안의 "오라클" 센서로 학습한 뒤 실제 촉각 데이터를 추론 시점에 정렬합니다. 제로샷 성공률 85~97.5%, 페그 삽입·기어 조립 과제에서 검증. (11/8)
https://www.alphaxiv.org/abs/2611.tactile-sim2real-latent-reconstruction
SBLR은 시뮬레이터 안의 "오라클" 센서로 학습한 뒤 실제 촉각 데이터를 추론 시점에 정렬합니다. 제로샷 성공률 85~97.5%, 페그 삽입·기어 조립 과제에서 검증. (11/8)
https://www.alphaxiv.org/abs/2611.tactile-sim2real-latent-reconstruction
🖐️ Michigan skips sensor-specific tactile simulation entirely
SBLR trains on a simulator-native "oracle" sensor, then aligns real tactile latents at inference. Zero-shot: 85-97.5% on peg insertion & gear meshing. (Nov 8)
https://www.alphaxiv.org/abs/2611.tactile-sim2real-latent-reconstruction
SBLR trains on a simulator-native "oracle" sensor, then aligns real tactile latents at inference. Zero-shot: 85-97.5% on peg insertion & gear meshing. (Nov 8)
https://www.alphaxiv.org/abs/2611.tactile-sim2real-latent-reconstruction
November 9, 2026 at 1:04 AM
🖐️ Michigan skips sensor-specific tactile simulation entirely
SBLR trains on a simulator-native "oracle" sensor, then aligns real tactile latents at inference. Zero-shot: 85-97.5% on peg insertion & gear meshing. (Nov 8)
https://www.alphaxiv.org/abs/2611.tactile-sim2real-latent-reconstruction
SBLR trains on a simulator-native "oracle" sensor, then aligns real tactile latents at inference. Zero-shot: 85-97.5% on peg insertion & gear meshing. (Nov 8)
https://www.alphaxiv.org/abs/2611.tactile-sim2real-latent-reconstruction
Aman Upganlawar: Geometric Refinement and Distributional Alignment for Sim2Real Roadside LiDAR Detection https://arxiv.org/abs/2611.03936 https://arxiv.org/pdf/2611.03936 https://arxiv.org/html/2611.03936
November 4, 2026 at 6:39 AM
Aman Upganlawar: Geometric Refinement and Distributional Alignment for Sim2Real Roadside LiDAR Detection https://arxiv.org/abs/2611.03936 https://arxiv.org/pdf/2611.03936 https://arxiv.org/html/2611.03936
Aman Upganlawar: Coordinate Frame Recovery, Per Class NMS and Null Return Modelling for Sim2Real V2X LiDAR Detection https://arxiv.org/abs/2611.03940 https://arxiv.org/pdf/2611.03940 https://arxiv.org/html/2611.03940
November 4, 2026 at 6:39 AM
Aman Upganlawar: Coordinate Frame Recovery, Per Class NMS and Null Return Modelling for Sim2Real V2X LiDAR Detection https://arxiv.org/abs/2611.03940 https://arxiv.org/pdf/2611.03940 https://arxiv.org/html/2611.03940
💫 pollen-robotics/microduck_rl — репозиторий с окружениями для обучения RL на бипедальном роботе Microduck.
• окружения построены на mujocolab (MuJoCo Warp) с PPO
• политики обучаются на частоте 50 Гц, экспортируются в ONNX и развёртываются на реальном роботе
• включает полный рецепт sim2real: BAM…
• окружения построены на mujocolab (MuJoCo Warp) с PPO
• политики обучаются на частоте 50 Гц, экспортируются в ONNX и развёртываются на реальном роботе
• включает полный рецепт sim2real: BAM…
October 27, 2026 at 4:26 PM
💫 pollen-robotics/microduck_rl — репозиторий с окружениями для обучения RL на бипедальном роботе Microduck.
• окружения построены на mujocolab (MuJoCo Warp) с PPO
• политики обучаются на частоте 50 Гц, экспортируются в ONNX и развёртываются на реальном роботе
• включает полный рецепт sim2real: BAM…
• окружения построены на mujocolab (MuJoCo Warp) с PPO
• политики обучаются на частоте 50 Гц, экспортируются в ONNX и развёртываются на реальном роботе
• включает полный рецепт sim2real: BAM…
4/6 Another approach to is do Sim2Real reinforcement learning, where the RL policy is trained in a simulation and then transferred to the real world. Despite impressive results in specialized domains, “it’s not going to get us to generalist robots,” said Bastani of @upenn.edu at TTIC.
October 22, 2026 at 9:13 PM
4/6 Another approach to is do Sim2Real reinforcement learning, where the RL policy is trained in a simulation and then transferred to the real world. Despite impressive results in specialized domains, “it’s not going to get us to generalist robots,” said Bastani of @upenn.edu at TTIC.
1/6 "We already have robots that can do really well on specialized tasks, but they struggle to do everyday tasks that humans take for granted in realistic, cluttered, real-world situations," said Osbert Bastani of @upenn.edu at his TTIC Colloquium talk. Video: www.youtube.com/watch?v=v54F...
"Scaling Sim2Real Reinforcement Learning" – Osbert Bastani, TTIC Colloquium
YouTube video by TTIC
www.youtube.com
October 22, 2026 at 9:13 PM
1/6 "We already have robots that can do really well on specialized tasks, but they struggle to do everyday tasks that humans take for granted in realistic, cluttered, real-world situations," said Osbert Bastani of @upenn.edu at his TTIC Colloquium talk. Video: www.youtube.com/watch?v=v54F...
"Scaling Sim2Real Reinforcement Learning," a TTIC Colloquium talk by Osbert Bastani of the University of Pennsylvania, is now available to stream on YouTube: buff.ly/jX2NJOD
"Scaling Sim2Real Reinforcement Learning" – Osbert Bastani, TTIC Colloquium
Enjoy the videos and music you love, upload original content, and share it all with friends, family, and the world on YouTube.
buff.ly
October 19, 2026 at 3:47 PM
"Scaling Sim2Real Reinforcement Learning," a TTIC Colloquium talk by Osbert Bastani of the University of Pennsylvania, is now available to stream on YouTube: buff.ly/jX2NJOD
But yeah, Spark might also work well for sim2real (although there as you said you need to actually run training on the server)
October 19, 2026 at 4:18 PM
But yeah, Spark might also work well for sim2real (although there as you said you need to actually run training on the server)
The other thing that blows my mind is that I haven’t come across anyone talking about sim2real for LM agents. In robotics it is helpful to train in simulation where accidents don’t have consequences. Then fine-tune on reality. Why are we not doing this for LM agents? Don’t tell me it’s too expensive
October 16, 2026 at 6:44 PM
The other thing that blows my mind is that I haven’t come across anyone talking about sim2real for LM agents. In robotics it is helpful to train in simulation where accidents don’t have consequences. Then fine-tune on reality. Why are we not doing this for LM agents? Don’t tell me it’s too expensive
松尾研究所と川崎重工が、ヒューマノイドロボットの「Kaleido9」を用いて、重い荷物の運搬デモに成功したようです。
NVIDIAのシミュレータを使うことで、わずか6週間で現実の世界での動作を実現したとのこと。
22.7キロもの重さがあるカゴを、両手でしっかりと運ぶ姿には、驚きを隠せません。
技術の進歩は、本当に目覚ましいものですね。
これからのロボットたちが、どのような未来を創り出すのか、とても楽しみです。
robotstart.info #news
NVIDIAのシミュレータを使うことで、わずか6週間で現実の世界での動作を実現したとのこと。
22.7キロもの重さがあるカゴを、両手でしっかりと運ぶ姿には、驚きを隠せません。
技術の進歩は、本当に目覚ましいものですね。
これからのロボットたちが、どのような未来を創り出すのか、とても楽しみです。
robotstart.info #news
ヒューマノイド「Kaleido9」の重量物運搬デモに成功した松尾研と川崎重工 NVIDIAシミュレータ活用により6週間でSim2Realを構築 | ロボスタ - ロボット･AI情報WEBマガジン
松尾研究所は川崎重工と連携し、ヒューマノイドロボット「Kaleido9」の強化学習Policyを共同開発した。NVIDIAシミュレータを使ったSim2Realを6週間で構築し、約22.7キロのカゴを両手で運ぶデモに成功した。
robotstart.info
October 14, 2026 at 3:04 AM
松尾研究所と川崎重工が、ヒューマノイドロボットの「Kaleido9」を用いて、重い荷物の運搬デモに成功したようです。
NVIDIAのシミュレータを使うことで、わずか6週間で現実の世界での動作を実現したとのこと。
22.7キロもの重さがあるカゴを、両手でしっかりと運ぶ姿には、驚きを隠せません。
技術の進歩は、本当に目覚ましいものですね。
これからのロボットたちが、どのような未来を創り出すのか、とても楽しみです。
robotstart.info #news
NVIDIAのシミュレータを使うことで、わずか6週間で現実の世界での動作を実現したとのこと。
22.7キロもの重さがあるカゴを、両手でしっかりと運ぶ姿には、驚きを隠せません。
技術の進歩は、本当に目覚ましいものですね。
これからのロボットたちが、どのような未来を創り出すのか、とても楽しみです。
robotstart.info #news
Built for an 800 g biped robot, microduckrl trains walking policies with MuJoCo Warp and PPO for direct sim2real transfer.
https://github.com/pollen-robotics/microduck_rl
Worth a look, or does it not survive contact with a real codebase?
https://github.com/pollen-robotics/microduck_rl
Worth a look, or does it not survive contact with a real codebase?
October 9, 2026 at 8:35 AM
Built for an 800 g biped robot, microduckrl trains walking policies with MuJoCo Warp and PPO for direct sim2real transfer.
https://github.com/pollen-robotics/microduck_rl
Worth a look, or does it not survive contact with a real codebase?
https://github.com/pollen-robotics/microduck_rl
Worth a look, or does it not survive contact with a real codebase?
Angel Wang, Dominique Perrault-Joncas, Alvaro Maggiar, Dean Foster, Carson Eisenach: Forecasting from Counterfactual Simulator Rollouts: A Sim2Real Evaluation https://arxiv.org/abs/2610.03662 https://arxiv.org/pdf/2610.03662 https://arxiv.org/html/2610.03662
October 5, 2026 at 6:45 AM
Angel Wang, Dominique Perrault-Joncas, Alvaro Maggiar, Dean Foster, Carson Eisenach: Forecasting from Counterfactual Simulator Rollouts: A Sim2Real Evaluation https://arxiv.org/abs/2610.03662 https://arxiv.org/pdf/2610.03662 https://arxiv.org/html/2610.03662
A new open-source ROS2 framework called multipanda_ros2 enables control of multiple Franka Robotics robots from a single process, sustaining a 1kHz control frequency with controller-switching delays of 2 ms or less. It also…
#ROS2 #robotics #sim2real #MuJoCo
https://arxiv.org/abs/2602.02269
#ROS2 #robotics #sim2real #MuJoCo
https://arxiv.org/abs/2602.02269
October 2, 2026 at 6:01 PM
A new open-source ROS2 framework called multipanda_ros2 enables control of multiple Franka Robotics robots from a single process, sustaining a 1kHz control frequency with controller-switching delays of 2 ms or less. It also…
#ROS2 #robotics #sim2real #MuJoCo
https://arxiv.org/abs/2602.02269
#ROS2 #robotics #sim2real #MuJoCo
https://arxiv.org/abs/2602.02269
RT @andrewmccalip: Do dumber things. I wired vision into JEV in a very hacky way and it worked great 😂
Last week’s rabbit hole was getting caught up on robot arms. Hardware, training pipelines, MuJoCo, sim2real, all of it.
Last week’s rabbit hole was getting caught up on robot arms. Hardware, training pipelines, MuJoCo, sim2real, all of it.
October 2, 2026 at 7:22 PM
RT @andrewmccalip: Do dumber things. I wired vision into JEV in a very hacky way and it worked great 😂
Last week’s rabbit hole was getting caught up on robot arms. Hardware, training pipelines, MuJoCo, sim2real, all of it.
Last week’s rabbit hole was getting caught up on robot arms. Hardware, training pipelines, MuJoCo, sim2real, all of it.
Somyajit Chakraborty, Xizhong Chen: Do Better Scores Mean Better Physics? Physics-Grounded Explanations for Sim2Real Neural Operators https://arxiv.org/abs/2610.00415 https://arxiv.org/pdf/2610.00415 https://arxiv.org/html/2610.00415
October 2, 2026 at 6:43 AM
Somyajit Chakraborty, Xizhong Chen: Do Better Scores Mean Better Physics? Physics-Grounded Explanations for Sim2Real Neural Operators https://arxiv.org/abs/2610.00415 https://arxiv.org/pdf/2610.00415 https://arxiv.org/html/2610.00415
He, Xie, Cao, Chen, Duan, Ge, Huo, Shao, Tian: Robot Manipulation with GPT-6-Astra: Body Knowledge, Experience Reuse, Emergent Skills, and Sim2Real Transfer https://arxiv.org/abs/2609.31770 https://arxiv.org/pdf/2609.31770 https://arxiv.org/html/2609.31770
September 29, 2026 at 6:44 AM
He, Xie, Cao, Chen, Duan, Ge, Huo, Shao, Tian: Robot Manipulation with GPT-6-Astra: Body Knowledge, Experience Reuse, Emergent Skills, and Sim2Real Transfer https://arxiv.org/abs/2609.31770 https://arxiv.org/pdf/2609.31770 https://arxiv.org/html/2609.31770
Monday, September 28 at 11:30am: TTIC Colloquium presents Osbert Bastani of UPenn with a talk titled "Scaling Sim2Real Reinforcement Learning." Please join us in Room 530, 5th floor.
September 25, 2026 at 5:15 PM
Monday, September 28 at 11:30am: TTIC Colloquium presents Osbert Bastani of UPenn with a talk titled "Scaling Sim2Real Reinforcement Learning." Please join us in Room 530, 5th floor.
NVIDIA Isaac Labを使って深層強化学習でオリジナル設計ヒューマノイドロボットをSim2Realで歩かせる
#ヒューマノイド #ロボット #深層強化学習 #IsaacLab #sim2real
#ヒューマノイド #ロボット #深層強化学習 #IsaacLab #sim2real
NVIDIA Isaac Labを使って深層強化学習でオリジナル設計ヒューマノイドロボットをSim2Realで歩かせる - Qiita
はじめに 自作の小型二足歩行ヒューマノイドロボット「スピロ」を、深層強化学習で歩かせた話です。 昨年はパラメータから足先軌道を生成する方式で歩かせていましたが、今年は歩容そのものを強化学習で獲得させ、学習済みポリシーをマイコン単体で動かすところまで持っていきました。この...
qiita.com
September 24, 2026 at 6:20 PM
#ROBOTIS launched the #OpenHumanoidGYMProject, an open-source initiative where #UniversityStudents train #humanoid #robots to perform advanced motions like backflips, K-pop choreography & obstacle running using simulation-to-real (#Sim2Real) AI testing. interestingengineering.com/ai-robotics/...
Video: Students push humanoid robots beyond walking with flips and more
ROBOTIS’ OH! GYM gives students hands-on experience building humanoid behaviors using AI Sapiens K1, simulation, and real-world testing.
interestingengineering.com
September 17, 2026 at 11:53 AM
#ROBOTIS launched the #OpenHumanoidGYMProject, an open-source initiative where #UniversityStudents train #humanoid #robots to perform advanced motions like backflips, K-pop choreography & obstacle running using simulation-to-real (#Sim2Real) AI testing. interestingengineering.com/ai-robotics/...
Struggling with bipedal sim2real transfer? Microduck RL trains locomotion policies in MuJoCo Warp and exports them to ONNX.
It has picked up 723 stars in the last week.
https://github.com/pollen-robotics/microduck_rl
It has picked up 723 stars in the last week.
https://github.com/pollen-robotics/microduck_rl
September 10, 2026 at 8:54 AM
Struggling with bipedal sim2real transfer? Microduck RL trains locomotion policies in MuJoCo Warp and exports them to ONNX.
It has picked up 723 stars in the last week.
https://github.com/pollen-robotics/microduck_rl
It has picked up 723 stars in the last week.
https://github.com/pollen-robotics/microduck_rl
At this year's Montreal Robotics Summer School, teams programmed a Unitree Go1 to walk itself to dinner — sim2real RL, vision-based nav, AprilTags and all. Congrats to our Winning teams (5 & 6)! Full recap 👉 fracturedplane.com/blog/2026/08/31/MRSS2026.html
#Robotics #ReinforcementLearning #Mila
#Robotics #ReinforcementLearning #Mila
September 8, 2026 at 6:41 PM
At this year's Montreal Robotics Summer School, teams programmed a Unitree Go1 to walk itself to dinner — sim2real RL, vision-based nav, AprilTags and all. Congrats to our Winning teams (5 & 6)! Full recap 👉 fracturedplane.com/blog/2026/08/31/MRSS2026.html
#Robotics #ReinforcementLearning #Mila
#Robotics #ReinforcementLearning #Mila
arXiv — Computers & Society + AI: ParaStudent: Closing the Sim2Real Gap in User Simulators for AI Tutor Evaluation
#AIinHigherEd #CsCY #CsAI
#AIinHigherEd #CsCY #CsAI
ParaStudent: Closing the Sim2Real Gap in User Simulators for AI Tutor Evaluation
Evaluating Artificial Intelligence (AI) tutor feedback before deployment requires anticipating student engagement, typically assessed through real interaction data. We introduce ParaStudent, a fine-tuning framework for simulating novice programming revisions to support AI tutor evaluation. Compared with prompted baselines, ParaStudent's revisions more closely match real student code distributions across functional, stylistic, and semantic metrics. Our best variant achieves AUCs of 0.80 for both feedback relevance and successful uptake when distinguishing streams with real engagement above versus at or below the median, while prompted baselines remain near chance on successful uptake. These findings demonstrate the promise of simulated engagement for pre-deployment feedback triage.
arxiv.org
September 4, 2026 at 4:20 PM
arXiv — Computers & Society + AI: ParaStudent: Closing the Sim2Real Gap in User Simulators for AI Tutor Evaluation
#AIinHigherEd #CsCY #CsAI
#AIinHigherEd #CsCY #CsAI
NVIDIA's Thomas Tang & iMerit's Vimal Patel discuss CV models moving from perception to reasoning, framed around the AI City Challenge; cross-city generalization + where anomaly reasoning falls short. Edge cases remain the hardest part: www.linkedin.com/feed/update/...
Booth #22, Malmö #ECCV2026
Booth #22, Malmö #ECCV2026
#computervision #autonomousdriving #sim2real #dataannotation #eccv2026 #aicitychallenge | iMerit Technology
NVIDIA’s Senior Deep Learning Engineer Zheng Tang spoke with iMerit's Vimal Patel to discuss how computer vision models are advancing from perception to reasoning, and where edge cases still challenge...
www.linkedin.com
September 3, 2026 at 2:18 PM
NVIDIA's Thomas Tang & iMerit's Vimal Patel discuss CV models moving from perception to reasoning, framed around the AI City Challenge; cross-city generalization + where anomaly reasoning falls short. Edge cases remain the hardest part: www.linkedin.com/feed/update/...
Booth #22, Malmö #ECCV2026
Booth #22, Malmö #ECCV2026
an argument I've heard for human-like hands is that you can train motion policies from video of human hands, and they transfer better
I'm not sure I fully buy this (I suspect sim2real gets better and hand-to-claw transfer is a thing), but like maybe as a hack to get things working faster?
I'm not sure I fully buy this (I suspect sim2real gets better and hand-to-claw transfer is a thing), but like maybe as a hack to get things working faster?
September 2, 2026 at 4:21 PM
an argument I've heard for human-like hands is that you can train motion policies from video of human hands, and they transfer better
I'm not sure I fully buy this (I suspect sim2real gets better and hand-to-claw transfer is a thing), but like maybe as a hack to get things working faster?
I'm not sure I fully buy this (I suspect sim2real gets better and hand-to-claw transfer is a thing), but like maybe as a hack to get things working faster?