MIM VLA uses motor feedback to guide robot grasping
A robot study tests physical feedback without adding tactile sensors.
A preprint posted October 6 introduces MIM VLA, which adds recent gripper motor readings to a robot learning system. It uses current, position and motion history to represent contact with an object.
The researchers report 75% accuracy when selecting the object with greater resistance across 13 pairs, compared with 48.8% for their SmolVLA baseline. Each pair received 20 trials per system.
Fragile objects remain difficult
The approach uses feedback already available from the gripper. The experiments do not establish a calibrated measurement of material hardness.
In a separate gentle grasping test, the system completed six of 20 grasps of an unseen raw egg without damage. Testing used one gripper platform. These are authors’ preprint results, with broader hardware testing and independent replication still needed.
ByteForward also covered Runway’s robot policy preview.
Illustrative robot workcell photograph by Nenad Stojković, also known as Shixart1985, under CC BY 2.0. Resized and converted to WebP. This unrelated hardware does not show the study’s setup.
