A robot that doesn't know what it's holding can't control it well. Mitsubishi's grant US12290928B2 ("Systems and methods for flexible robotic manipulation by fast online load estimation," issued May 6, 2025; inventors Yebin Wang, Xiaoming Duan, Toshiaki Koike-Akino, Philip Orlik) fences the fix: estimate the payload's mass and dynamics quickly, online, while the robot is already manipulating it — then control accordingly.
The mechanism is fast load identification feeding force control: B25J 9/1633 (control with force/torque feedback) is the operative code, and the 'online' qualifier means the estimate happens during motion, not in a separate calibration step. For flexible manipulation of varied, unknown objects — a warehouse or a kitchen, not a fixed assembly line — knowing the load lets the controller compensate for inertia and gravity it couldn't otherwise predict.
For the manipulation beat, online load estimation is the control-side complement to the sensing-side work (slip detection, tactile arrays) that other players fence. All of it addresses the same reality: real manipulation deals with objects whose properties you don't know in advance, so the robot has to identify them in the loop. A claim on fast online estimation fences the adaptation that makes general-purpose manipulation controllable.
From a portfolio angle, this extends MERL's manipulation-research line (the 2021 dexterous-gripper grant US11104008B2 sits in the same portfolio). MERL files research-grade control-and-mechanism patents that seed Mitsubishi's industrial-automation products and licensing. An adaptive-control grant like this is the algorithmic counterpart to its earlier mechanism IP — building breadth across both the hand and its control.
Caveats. Online parameter and load estimation has a long adaptive-control prior-art history; the grant turns on the specific fast-estimation method in claim 1, not on the concept of identifying a load. The two-code classification marks it as force-control manipulation without specifying the estimator. Read the independent claim for the estimation algorithm.
For the file: a fast online load-estimation grant on the control side of adaptive manipulation, extending MERL's portfolio. Pull US12290928B2 (and the 2021 gripper sibling) on PatentBear, read claim 1 for the estimation method, and file it as the control complement to tactile-sensing manipulation IP.