Suraj Nair

As in-hand manipulation is gaining traction, it is more important than ever to compare and contrast our progress. A diverse group came together to define a protocol for Benchmarking In-Hand Manipulation -- https://arxiv.org/pdf/2001.03070.pdf …pic.twitter.com/JKB9ow0pwz

We present Interactive Gibson Benchmark, a benchmark for interactive navigation in cluttered environments. We provide baselines with different trade-offs between navigation path efficiency and disturbance of surrounding objects. Code + paper: https://sites.google.com/view/interactivegibsonenv …pic.twitter.com/7ZYsF9XGIK

Excited to share PCGrad, a super simple & effective method for multi-task learning & multi-task RL: project conflicting gradients On Meta-World MT50, PCGrad can solve *2x* more tasks than prior methods https://arxiv.org/abs/2001.06782  w/ Tianhe Yu, S Kumar, Gupta, @svlevine, @hausman_kpic.twitter.com/uTeUhULUTA

We propose a method for in-hand manipulation that combines low-level manipulation primitives with a learned, mid-level policy that orchestrates these primitives. Our method keep the object firmly grasped while transporting it to distant goal poses. https://sites.google.com/view/learninghierarchicalcontrol …pic.twitter.com/sH6lRQxOfi

We are happy to introduce Action Genome: a new representation, new dataset, and new model for decomposing actions into spatio-temporal scene graphs. Action Genome has 1.7M relationships between 0.4M object instances and enables few-shot action prediction. https://arxiv.org/pdf/1912.06992.pdf …pic.twitter.com/xU774HLX3O

Can we discover structure & meta-learn across it in unsegmented time series data? MOCA simultaneously detects changepoints & meta-learns across time for continuous adaptation Continuous Meta-Learning without Tasks https://arxiv.org/abs/1912.08866  w @jmes_harrison, Sharma, @MarcoPavoneSUpic.twitter.com/fDMCd9MumV

We're releasing Procgen Benchmark, 16 procedurally-generated environments for measuring how quickly a reinforcement learning agent learns generalizable skills. This has become the standard research platform used by the OpenAI RL team: https://openai.com/blog/procgen-benchmark/ …pic.twitter.com/OhECCCAeY3

Sudeep Dasari wrote an excellent blog post on our work on RoboNet, cross-posted on the SAIL blog: http://sail.stanford.edu/blog/robonet/  We imagine a future where robots share data *across* research labs, just like the rest of machine learning. https://twitter.com/svlevine/status/1199220226807590912 …pic.twitter.com/c77Sxl0Vqt

We're releasing Safety Gym, environments and tools to evaluate reinforcement learning with safety constraints: https://openai.com/blog/safety-gym/ … Aims to ultimately help agents satisfy real-world safety requirements while training (eg not driving off a cliff, not writing abusive content).pic.twitter.com/VTwS4KoFS1

We beta-released the IKEA Furniture Assembly environment for solving long-term manipulation tasks with high-quality rendering. * 80+ furniture models * Multiple robots * Various observations * Supports OpenAI gym Try it out and let us know what you think! https://clvrai.com/furniture pic.twitter.com/1id4EWeqzu