Register and share your invite link to earn from video plays and referrals.

HRX
@Spaulhrx
Joined July 2024
0 Following    0 Followers
đ—Ŗđ—Ĩ𝗜đ—Ļ𝗠𝗔đ—Ģ: đ—Ē𝗛𝗘𝗡 đ—Ļ𝗜𝗠𝗨𝗟𝗔𝗧𝗜đ—ĸ𝗡 𝗠𝗘𝗘𝗧đ—Ļ đ—Ĩ𝗘𝗔𝗟𝗜𝗧đ—Ŧ A robot can perform well in simulation and still struggle with a simple task in the real world. That's something I've been exploring while following @PrismaXai . At IROS 2026 in Pittsburgh, PrismaX demonstrated its work on robotic manipulation, using a task as simple as picking up fruit and placing it into containers. But the interesting part is what happened before the demonstration. đ—ĸ𝟭 / đ—ĸ𝘃𝗲đ—ŋ 𝟭đŸŦ,đŸŦđŸŦđŸŦ đ—Ļđ—ļđ—ē𝘂𝗹𝗮𝘁𝗲𝗱 𝗧𝗲𝘀𝘁𝘀 PrismaX ran thousands of simulated tests to understand where its robotic system might fail. The goal wasn't just to make the robot complete a task once. It was to identify weaknesses before testing on real hardware. đ—ĸ𝟮 / 𝗧đ—ŋ𝗮đ—ļđ—ģđ—ļđ—ģ𝗴 𝗠𝗮𝘁𝘁𝗲đ—ŋ𝘀 One comparison from their reported experiments stood out. → 51% pickup success when training only the action component. → 89% when training the full model, including the vision and language backbone. That suggests the robot benefits from adapting more than just its movement predictions. The tradeoff? Training the full model takes more time. đ—ĸđŸ¯ / đ—Ĩ𝗲𝗮𝗹 đ—Ēđ—ŧđ—ŋ𝗹𝗱 đ—Ŗđ—˛đ—ŋđ—ŗđ—ŧđ—ŋđ—ē𝗮đ—ģ𝗰𝗲 Even with promising simulation results, the real robot didn't perform as well as expected. The team also investigated inference latency and reported optimizations that reduced it substantially. It's a useful reminder that robotics isn't only about having a capable AI model. Hardware, timing, data quality, and the physical environment all matter. 𝗠𝘆 𝗧𝗮𝗸𝗲 What interests me about PrismaX is its focus on the practical challenges of physical AI. Getting a robot to understand an instruction is one thing. Getting it to execute that instruction reliably, on real hardware, under real-world conditions, is a completely different challenge. And I think that's where some of the most important work in robotics is happening. This demonstration is one specific task, with results reported by PrismaX rather than independent benchmarks. But the lessons about simulation, training, and real world execution are worth understanding. The future of physical AI won't be decided by impressive demos alone. It'll depend on how consistently robots can perform when conditions aren't perfect.
Show more
Robotics is advancing quickly but there's still a big gap between what robots can learn in training and what they can actually do in the real world. That's one reason I've been following @PrismaXai They are working on connecting human teleoperation, real world data collection, and robotics training to help physical AI improve through practical experience I wanted to explore that idea visually so I put together this short animation about what PrismaX is building Hope you enjoy watching it as much as I enjoyed creating it.
Show more