sim2real

Perspectives on Sim2Real Transfer for Robotics: A Summary of the R:SS 2020 Workshop arxiv.org/abs/2012.03806

ProcThor - procedural generation of interactive and diverse 3D environments to train simulated embodied agents. arxiv.org/abs/2206.06994 procthor.allenai.org

reminder that the paper submission deadline is just one week away i.e. may 31st, 2022

The third workshop on sim2real transfer for robotics is happening at R:SS 2022 at New York City on June 27th. Please consider submitting your recent work on sim2real here. sim2real.github.io

Bi-Manual Manipulation and Attachment via Sim-to-Real Reinforcement Learning arxiv.org/abs/2203.08277

We introduce a new method to learn simulators from depth and RGB videos. The "URDF" of an articulated rigid-body mechanism is reconstructed, and the parameters of the simulator inferred through Bayesian inference. Website: eric-heiden.github.io/video2sim/ Paper: arxiv.org/abs/2203.10488

We have released a super fast implementation of AMP using @nvidia's Isaac Gym. Models that used to take days to train can now be trained in minutes! github.com/NVIDIA-Omniver… There's also a retargeting script for loading new motion clips. Thanks to @viktor_m81, @gavrielstate.

CenterSnap: Single-Shot Multi-Object 3D Shape Reconstruction and Categorical 6D Pose and Size Estimation arxiv.org/abs/2203.01929 zubair-irshad.github.io/projects/Cente…

Super excited to share my internship work at @ToyotaResearch on category-level 3D object understanding and single-shot real2sim asset creation, accepted at #ICRA2022! zubair-irshad.github.io/projects/Cente… @GTrobotics @ICatGT @mlatgt @ieee_ras_icra ⬇️(1/6)

Kubric: A Scalable Dataset Generator arxiv.org/abs/2203.03570

Dojo: A Differentiable Simulator for Robotics arxiv.org/abs/2203.00806

We at @StandardAI are thrilled to introduce StandardSim, a Photorealistic Synthetic #dataset for #AI research in Retail Environments! 25K+ high-resolution multi-view renders with pixel-perfect labels 🚀 standard-ai.github.io/Standard-Sim/ arxiv.org/abs/2202.02418

We are releasing MOTSynth, the largest dataset for Pedestrian Detection and Tracking. Expect to obtain better real-world performances by training your model on this data only. It's synthetic, but it's for real! lnkd.in/d-NvqiX4

Happy to announce that the 3rd Embodied AI Workshop is happening on June 19th at #CVPR2022. Lots of exciting speakers and 13 (!!!) challenges. Check out our website (embodied-ai.org), dive into a challenge, and save the date!

Released our work on object extraction: SafePicking #ICRA2022! This robotic system finds and “safely” extracts target objects via known object mapping with an onboard camera. Check out the project page for paper/code: safepicking.wkentaro.com Work with @stepjamUK @AjdDavison (1/n)

What I’ve learned about making synthetic data work for training ML models:

DROPO: Sim-to-Real Transfer with Offline Domain Randomization arxiv.org/abs/2201.08434

Some simulation platforms that caught my eye this year, 2021 🧶

Realistic and diverse simulation is a crucial tool for building AV2.0 - safe, scalable, and adaptable learned driving. If you share our passion for sim and AV, and want to work in a dynamic and diverse team at the forefront of AI, get in touch at careers@wayve.ai @wayve_ai.

Learning Vision-Guided Quadrupedal Locomotion end-to-end with Cross-Modal Transformers arxiv.org/abs/2107.03996 rchalyang.github.io/LocoTransforme…

Need Sim2Real baselines? On finishing his PhD with us, Fabio has released version 1.0 of SimuRLacra github.com/famura/SimuRLa… a PyTorch-based RL framework focusing on Sim2Real transfer methods