← Paper Digest

Natural Swing-Up of an Underpowered Double Pendulum

Arne Sachtler, Manuel Keppler, Florian Loeffl, Alin Albu-Schäffer

November 2026

TL;DR A gain-free controller swings up a direct-drive two-link robot whose motors give only 6% of its shoulder gravity torque. It splits torque into work-free steering and energy pumping, chosen by a two-variable linear program. On hardware it reaches near-upright energy in 13.6 s and 7.7 s along the robot's two nonlinear normal modes, staying closer to each mode than two baselines.

What they do

The direct-drive motors of this planar two-link robot deliver 1.1 Nm, about 6% of the shoulder’s gravity torque. At full torque, link 2 tilts only 16.3°, so the robot must build up energy over many swings. The authors pump energy along its nonlinear normal modes. These are families of periodic orbits of the unforced dynamics that lead from hanging to almost upright, where a PD controller catches the robot. Both modes are non-strict: adding energy pushes the state off the mode, so pumping and stabilization compete for the same bounded torque.

How it works

  • Power-based split. Torque is divided into steering, which does no work, and pumping along the momentum, the only part that changes the energy. A feedforward term cancels friction.
  • Two-variable LP. At 1 kHz, the controller maximizes pumping subject to per-joint torque limits, non-negative pumping (energy never drops), and a soft control-Lyapunov constraint that makes the error to the mode decay at rate ρ. The LP is solved in closed form.
  • No feedback gains. ρ is set about 20% above the transverse instability (Floquet exponents) of the mode’s orbits.

Results

  • Hardware swing-ups succeed on both modes, reaching 33 J (upright: 34.25 J) after 13.6 s and 7.7 s, then the robot is caught upright.
  • Against prioritized stabilization on mode 1: 13.6 vs. 13.7 s, with median distance from the mode 0.05 vs. 0.08. The baseline needed a hand-tuned gain schedule. Mode 2: 8.0 vs. 8.8 s, with distance 0.07 vs. 0.39.
  • Summing and clipping took 17.4 s on mode 1 and never worked on mode 2, where its energy peaked at 26 J.

Limitations

Shown on two joints only. Near turning points neither input has authority, and no analysis covers that band. The catch needs a separate controller, the allocation is greedy rather than time-optimal, and the CLF constraint can be infeasible (up to 10.8% of cycles). The paper reports individual runs and no success rate over repeated trials.

Figures

From Figure 1: The robot hardware, a planar two-link pendulum with direct-drive joints.
From Figure 1: The robot hardware, a planar two-link pendulum with direct-drive joints.
From Figure 4: CAD model of the robot: shoulder motor (left), link 1 with the electronics, elbow motor, and link 2.
From Figure 4: CAD model of the robot: shoulder motor (left), link 1 with the electronics, elbow motor, and link 2.
From Figure 6: The three controllers (sum and clip, prioritized, ours) from rest to 33 J on both modes. Left: energy ...
From Figure 6: The three controllers (sum and clip, prioritized, ours) from rest to 33 J on both modes. Left: energy over time. Right: median Lyapunov error W, the distance from the mode, per 1 J of energy.
Abstract

Robots with weak motors must ride their natural dynamics rather than fight them. We swing up a direct-drive planar two-link robot, whose motors deliver only about 6 % of its shoulder gravity torque, along its nonlinear normal modes, two-dimensional invariant manifolds consisting of periodic orbits. On a non-strict mode, injecting energy necessarily moves the state off the manifold, so pumping and stabilization compete for the same bounded torque. We split the input torque into a power-neutral steering component and a scalar pumping input, the only component that changes the energy. The allocation then becomes a linear program in two variables: maximize pumping subject to the torque limits and one control-Lyapunov constraint. The constraint uses the model-based rate of the error, so no stabilizing feedback gain is needed. Experiments in hardware show swing-ups along both modes and compare the allocation with summing-and-clipping and with prioritized stabilization.

Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.