Natural Swing-Up of an Underpowered Double Pendulum
Authors: Arne Sachtler, Manuel Keppler, Florian Loeffl, Alin Albu-Schäffer
Organizations: Technical University of Munich (TUM), Munich, Germany. · German Aerospace Center (DLR), Institute of Robotics and Mechatronics, Oberpfaffenhofen, Germany. · Eindhoven University of Technology (TU/e), Eindhoven, The Netherlands.
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.
Figures & tables
Fig. 1: (a) Close-up of the robot, link 1 hanging and link 2 upright: direct-drive motors in both joints, electronics and battery on the links. (b) With full torque in both joints, the model’s static pose is q=(0∘,16.3∘) : link 1 hangs straight down and link 2 tilts by 16.3∘ . (c) Model: link lengths li , centers of mass ri , masses mi , inertias Ii , viscous damping di ; q1 from the downward vertical, q2 relative to link 1.
Fig. 2: (a) Brake orbits of NNM 1 and NNM 2 at 2 to 34 J (light to dark) on the configuration torus Q=S1×S1 , drawn as a donut (NNM 1 left, NNM 2 right, also in (b)); q1 runs around the large circle, q2 around the tube, and q2=0 is the outer equator (grey), with the hanging point at the front; arrows: increasing q1 , q2 ; fine dashes: parts behind the surface. (b) The same torus in the kinetic-energy metric, embedded isometrically as a surface of revolution, one copy q2∈[−π,π] ; grey: lines q1=const . Dashed red in (a) and (b): the cut q2=±π , red dot: (q1,q2)=(π,±π) ; in (b) the two rims are the same circle, identified after a rotation by 54.7∘ . (c) Period T and Floquet exponent λF of the free orbits over their energy; dashed: Eupup .
measure
NNM 1
NNM 2
orbit tangent at q=0 vs. linear mode, Euclidean
45∘
5∘
same, KE metric M(0)
14.7∘
23.0∘
s ( 3 ), mean over φ and E
0.11
0.16
s , peak over E of the mean over φ
0.26
0.29
h/L , Euclidean q
3.6 %
2.7 %
h/L , KE metric
0.9 %
3.6 %
TABLE I: Non-strictness of both modes. The ranking flips between Euclidean q and the kinetic-energy (KE) metric.
Fig. 3: (a) The split in kinetic-energy coordinates v=M1/2q˙ : the pumping input γ stretches the velocity, the steering β rotates it at constant kinetic energy (circle). (b) The program ( 11 ) in the u=(β,γ) plane at one state of a simulated swing-up along NNM 1 ( E=15.0 J): box constraint of both joints, bounds on γ (dashed), CLF constraint (orange), U(z) (light), feasible set (dark), optimum and tie interval. The CLF constraint is active: the error, not the box, limits the pumping.
Fig. 4: CAD model of the robot: shoulder motor (left), link 1 with the electronics, elbow motor, and link 2.
parameter
link 1
link 2
unit
length li
0.5
0.408
m
mass mi
3.143
1.078
kg
center of mass ri
0.257
0.370
m
inertia Ii
0.135
0.008
kg m 2
damping di
0.018
0.007
Nms/rad
torque limit τmax
1.1
1.1
Nm
TABLE II: Robot parameters (identified).
Fig. 5: Swing-up along NNM 1 and NNM 2 with the same controller, each with its rate from ( 12 ). (a) The robot upright. (b), (c) Path of the tip from the kick to the end of the run, shaded by energy, with the robot at turning points and the final upright pose after the catch (black). (d) Energy E , Lyapunov function W and pumping input γ of both runs.
Fig. 6: The three controllers from rest to E=33 J on both modes. Left: energy. Right: median of W ( 4 ) per 1 J of energy.
method
t(E=33J) [s]
d [ J ]
infeas.
NNM 1
sum and clip ( V-C )
17.4
0.43 / 1.06
n/a
prioritized ( V-C )
13.7
0.08 / 0.19
n/a
ours
13.6
0.05 / 0.10
1.1 %
NNM 2
sum and clip ( V-C )
–
–
n/a
TABLE III: Comparison with the baselines (–: 33 J not reached)
The 4th "AI Olympics with RealAIGym" competition, to be held at IJCAI-ECAI 2026 in Bremen, challenges participants to develop a global control policy for swinging up and stabilizing an underactuated two-link system in its upright position. In contrast to previous editions, participants develop and evaluate their control strategies directly on remotely accessible CloudPendulum hardware, with limited interaction time and without prior knowledge of the system's model parameters. This paper presents an optimal-control-based approach employing real-time nonlinear model predictive control implemented using sequential quadratic programming. The results demonstrate that the proposed SQP-based MPC controller achieves reliable swing-up and stabilization performance, while maintaining robustness against disturbances.
Nick Karydakis, Konstantinos Chatzilygeroudis
Laboratory of Automation and Robotics (LAR) in the Department of Electrical & Computer Engineering, University of Patras, GR-26504 Patras, Greece
The cart-pole swing-up is a canonical benchmark for nonlinear control of underactuated systems, yet an end-to-end guarantee linking the global swing-up maneuver to the local stabilizer is seldom formalized. We present a reachability analysis of a switched energy-based/LQR controller that certifies convergence to the upright equilibrium from a compact set of initial conditions. The swing-up law is derived from an energy-error Lyapunov function; canceling the autonomous conservative term yields a strictly sign-definite Lyapunov derivative, and convergence follows from LaSalle's invariance principle. We also propose an augmented Lyapunov function to regulate the steady-state cart velocity to zero, for which we establish almost-global convergence. For the controller handoff, a switching region is designed to lie strictly within the LQR region of attraction, formally certifying the swing-up-to-stabilization transition. Numerical simulations corroborate the theoretical analysis.
Mohamed Khalid M Jaffar
Department of Aerospace Engineering University of Maryland, College Park, USA
Soft and soft-rigid hybrid robots are inherently underactuated and operate under tight actuator limits, making task-space control with stability guarantees challenging. Common nonlinear strategies for soft robots (e.g., those based on PD control) often rely on the assumption of full actuation with no actuator limits. This paper presents a general control framework for task-space regulation and tracking of underactuated soft robots under bounded inputs. The method enforces a rapidly exponentially stabilizing control Lyapunov function as a convex inequality constraint while simultaneously satisfying underactuated full-body dynamics and actuator bounds. We validate the approach in simulation on several platforms spanning increasing underactuation: a simple two link tendon-driven "finger", a trimmed helicoid manipulator, and a highly underactuated spiral robot. We compare against a number of baseline methods from the literature. Results show improved task-space accuracy and consistent Lyapunov convergence under input limits, achieving superior set-point and trajectory-tracking performance.
Huy Pham, Zach J. Patterson
Mechanical and Aerospace Engineering, Case Western Reserve University