详细解读

Robots ca not learn from simulation alone they need real world data. Thats exactly what PrismaX is working on with UPenn GRASP Lab providing real world robotics data to support robot learning research through its University Research Program. UPenn is the first publicly announced university partner in this program

The capstone team consists of four Robotics master students Kaifan Yu, Sumukh Shrote, Smrithi Lokesh, and Dayoon Suh under the supervision of Professor Antonio Loquercio. Their research explores how reinforcement learning can help robotic policies adapt to tasks and environments they have never encountered during training

The baseline has already been completed on LIBERO PRO. Using RLDT, a method from Professor Loquercio’s previous research, the team increased the success rate from 40% to 78% in one object layout variation scenario. The next step is to incorporate YAM demonstrations and compare supervised fine tuning (SFT) alone against SFT followed by reinforcement learning

PrismaX is providing the research team with:
YAM teleoperation episodes from its existing data bank
Pick and place tasks involving fruit, toothbrushes, and first aid kits
Access to existing datasets or the option to request custom data collection based on thesis specifications
Free data access for selected projects from accredited universities and research laboratories

The collaboration was formalized at the Fall 2026 GRASP Lab Industry Day on October 9 at the Singh Center for Nanotechnology
Shaye Backus, Head of Growth at @PrismaXai presented during the morning industry talks
Vivian Li, Marketing Operations Lead, introduced the University Research Program
The team connected with Professor Vijay Kumar, Professor Camillo J. Taylor, and Professor Ani Hsieh during the faculty lunch and evening reception
A week earlier, they had also met at IROS 2026 in Pittsburgh, where GRASP had a major presence

Understanding PrismaX and GRASP Lab
PrismaX is a robotics native service layer for Physical AI, backed by a16z and NVIDIA Inception
Its offering covers the full stack diverse, usable datasets across teleoperation and egocentric modalities, alongside end to end software that enables a decentralized network of owner operators and reviewers to capture and verify data while getting paid for their contributions

Meanwhile, founded in 1979, UPenn’s GRASP Lab is one of the world's most respected robotics research centers. It serves as a leading hub for research in vision, perception, control, automation, and machine learning within Penn Engineering

UPenn is just one of the first university partnerships PrismaX is building, and the opportunity remains open
If your lab needs realworld robotics data, whether from an existing data bank or through custom collection, you can apply to the PrismaX University Research Program. Selected projects receive access to the data for free

The future of robotics is being trained with the data we collect today, and that data is now becoming accessible to research labs everywhere