Skild AI's Steel: 8× More Parallel Industrial Robot Cells in Our Benchmark
About SN58 Attune
Robotensor@robotensor_aiSkild AI's Steel: 8× More Parallel Industrial Robot Cells in Our Benchmark We integrated @SkildAI's newly open-sourced Steel physics solver into Isaac Lab 3.0 and benchmarked it against PhysX on an RTX PRO 6000 Blackwell. Skild AI reports 3.8× higher simulation throughput across 20 single-task benchmarks and up to 13× on heterogeneous workloads. Our experiments independently confirm Steel's impressive performance and scalability. 1. Steel outperformed PhysX across all five standard tasks. At 4,096 parallel environments: Ant: 4.6× UR10 Reach: 3.4× Cartpole: 2.3× Franka Reach: 2.3× Humanoid: 1.8× A 2.7× geometric-mean speedup, with a 2.3× advantage maintained at 16,384 environments. 2. Steel demonstrated exceptional scalability in industrial machine tending. We simulated a UR10e, Robotiq 2F-85 gripper, 7-axis machine, and 30 industrial parts. PhysX handled 512 parallel cells but crashed at 1,024. Steel successfully ran 4,096 cells on the same GPU, with 98% successful placements. That's 8× more parallel robot cells than PhysX could sustain in our setup. 3. Contact physics remains an opportunity for improvement. Robot trajectories agreed within 1.2 mm, but Steel tilted a clamped part by up to 36°, compared with PhysX keeping it flat. Both met our 12 mm placement tolerance (Steel: 6 mm; PhysX: 1 mm). PhysX also delivered higher per-cell simulation throughput in this particular environment. Stiff-contact tuning and missing contact-force sensor support remain limitations for our manipulation workloads. Our takeaway: Steel is a remarkable open-source contribution from Skild AI and NVIDIA. Its efficiency on standard tasks and ability to scale complex industrial simulations are especially promising. Improving contact-heavy manipulation could make it even more valuable for industrial robot learning. Huge thanks to @deepakpathak and @SkildAI for open-sourcing this work. We're excited to build on Steel and share our experimental findings with the community.…
Linkskild.aiskild.ai/blogs/steelSN58 Attune at a glance
SN58, Attune (α token خ), is one of the independent markets on the Bittensor network, each with its own α token and its own price. Its α currently trades at 0.013648 τ (≈ $3.06), against 1,739 τ of pool liquidity. On-chain, the best-delegated validator on it yields about 75.93% APR (simple) and 113.67% APY (compounded), emissions add 0.006458 τ to the pool each block, epochs run every 360 blocks. It was registered at block #8,511,017.
Taosis newsletters
Taosis News: the week’s articles from the subnet teams, the press and community newsletters, every Friday. Taosis Weekly: the week’s movers, best yields and chain events, with dollar values, every Monday. Past issues
More from SN58 Attune
Taosis indexes announcements, articles, posts and releases about Bittensor subnets from their own channels and keeps a record of each. The text above is the source’s; the figures are read from the chain by Taosis.