ASHIK

ASHIK

Users
Tweets

ASHIK

@Ashik_80

May 25

If your @axisrobotics task failed recently — you're not alone. A lot of contributors are confused about why it happened and what to do next. Here's the reasons: → Bot activity — Task 805 had unusually high failure rates because bots were spamming it, not real players → Gripping & manipulation issues — failed grasps, objects slipping, gripper problems mid-task → Physics glitches in sim — lag, objects clipping, unstable long-task sessions → Harder new tasks — microwaves, ovens, cabinets require open/close/press skills that are genuinely more complex So what should you do? Honestly? Just move on to the next task. You can't fix the bots, the sim glitches, or the data pipeline yourself. Don't stress over a failed task — it's not always on you. #AxisRobotics #PhysicalAI #RoboticsAI #TaskFailure #SimToReal #AIContributors

Axis Robotics

@axisrobotics

May 25

Axis Weekly This week, we focused on making the robotics data loop more measurable and reproducible: separating real user signals from bot traffic, expanding TaskGen into articulated-object tasks, and turning data-to-model workflows into repeatable services. Key updates: - Data quality: Task 805’s high failure rate was driven by bots, not real players. - TaskGen: Codebase delivered for an upcoming update that will support end-to-end generation of articulated-object tasks from prompts. - Simulation and data infra: Asset bugs fixed, and the automated recover-from-failure pipeline is nearing full deployment. - Model training: Achieved a ~40% success rate in cross-simulation evaluation (IsaacLab to MuJoCo). - Sim-to-real: Updated the domain randomization roadmap to heavily boost physical parameter diversity. A closer look at this week’s progress 🧵

0:06