A flexible robot arm that bends like an elephant’s trunk can slip into tight spaces and handle fragile objects, but its smooth bending makes its next move hard to predict. Researchers at Peking University have developed a hybrid model that keeps computer forecasts of a flexible robot arm’s motion on track, even when the arm moves faster and harder than the model was trained for. The work was described in a Peking University story carried by TechXplore on 23 October 2026.
The study involves professors Xie Guangming and Zhang Feitian at Peking University’s School of Advanced Manufacturing and Robotics, with Ahsan Tanveer as first author. It was presented at Robotics: Science and Systems (RSS 2026), held in Sydney from 13 to 17 July 2026, and is published in the conference proceedings, Robotics: Science and Systems XXII.
Why a flexible robot arm is hard to forecast
Before a flexible robot moves, its controller often rehearses possible commands in a computer model and picks the best one. Models trained on data predict one small step at a time, and each step starts from the last, so small errors pile up. After 200 steps, the forecast can be far off, and the robot can end up working from a false picture of its own body.
How the hybrid flexible robot arm model works
The team’s idea is to give the forecast a reference point. The first part, called the anchor, learns the shape the arm settles into whenever its controls are held steady. It was fitted from 4,520 static equilibrium samples. At every step, the anchor pulls the prediction back toward that resting shape, so errors cannot snowball.
The arm, however, swings and bends on its way to rest. A second, lightweight model, the correction, learns those movements from the arm’s motion data. The researchers say their method separates steady-state anchoring from transient correction at every step.
Flexible robot arm test results
The team first tested the idea in simulation, on a three-dimensional arm driven by cables. The model was asked to predict 6 seconds of motion, or 200 steps, under stronger and faster actuation than it saw in training. Across 19 test trajectories, the hybrid model cut tip position error by 27% and backbone position error by 26% compared with the anchor alone. It never lost track in any of the 19 tests, while the correction alone could drift into physically impossible shapes.
The researchers then moved to real hardware: a soft fishlike tail that swings from side to side as it is pressurised with water. When the swing frequency shifted, the hybrid model’s error was half or less that of the anchor alone and a quarter or less that of the correction alone. On standard benchmark tests it finished every prediction with no catastrophic failures, while a competing Koopman-based method failed on harder starting conditions.
What steadier predictions could enable
Steadier predictions give a robot’s controller something it can trust. According to the release, a controller could try out many possible moves on the computer first and choose the safest one before the real arm moves. That could help flexible arms work in tight spaces where rigid machines struggle, and with objects that break easily. For another take on soft, sensitive robot arms, see our story on HKUST’s whole-arm tactile skin.
The case for and against
The case for this work is that it tackles a basic weakness of learned models: errors that pile up over long forecasts. Using cheap static data as an anchor is a simple, practical idea.
The case for caution is that the 3D arm results come from simulation. The real-world test used a soft tail rather than a full 3D arm, and the release does not describe the model controlling a robot in a real task. The research was also first presented in July.
My view on the flexible robot arm research
As a writer, I feel this is the kind of quiet, foundational work that soft robotics needs. Flexible arms get attention for their looks, but they are only useful if their controllers can trust their own forecasts. I would like to see the method running on a physical 3D arm doing a real job next.
FAQ: Peking University flexible robot arm
What problem does the Peking University flexible robot arm model solve?
It stops small step-by-step prediction errors from piling up, keeping long forecasts of the arm’s motion on track.
How much did the hybrid model improve predictions?
In simulation, it cut tip position error by 27% over 200-step forecasts and stayed stable in all 19 tests.
Where was the research presented?
At Robotics: Science and Systems (RSS 2026) in Sydney, held from 13 to 17 July 2026.