Central pivot point control of a body-attached spring-mass runner: periodic gaits, stabilization, template embedding and approximate analytic return maps
Authors: İzel Sever, O. Kaan Karagöz, M. Mert Ankaralı
Organizations: Department of Electrical and Electronics Engineering, Middle East Technical University, 06800 Ankara, Türkiye · Roketsan Inc., Ankara, Türkiye · Center for Robotics and Artificial Intelligence (ROMER), Middle East Technical University, 06800 Ankara, Türkiye
The spring-loaded inverted pendulum (SLIP) describes the center-of-mass (COM) motion of runners but has no trunk. In trunk models, the hip torque couples posture and COM motion and destroys the central-force structure on which SLIP analysis relies. As a result, trunk models have been studied almost only numerically. We show that this structure can be restored. The central pivot point (CPP), a hip-torque law that directs the ground reaction force through the COM, decouples pitch exactly and reduces stance to a planar central-force problem. At upright posture, time-reversal symmetry produces one-parameter families of periodic gaits. A stance-phase pitch law and a stride-level regulator stabilize all computed gaits, also with bounded hip torque and friction. Under CPP, any radial force law in virtual-leg coordinates can be embedded independently of posture. For a linear radial template, the stance radius and the constant-momentum sweep have exact closed forms. With a new gravity correction, the resulting analytic return map, to our knowledge the first for a trunk model, predicts liftoff velocity and apex position with mean errors of 1.6% and 0.9%. The natural gaits of the embedded runner are symmetric and need no force step at touchdown. A near-reversibility argument explains their unit multiplier, and a scalar equation predicts them within 0.8%. A deadbeat stride controller based on the analytic map tracks changes of speed, height and posture. CPP thus makes trunk models accessible to the analytic tools developed for SLIP.
Figures & tables
Figure 1: (a) Trunk-SLIP (TSLIP) model and its parameters; the labels dhip and d correspond to dh and c in the text. (b) Central pivot point (CPP): the hip torque τ adds a force component Fτ=τ/r normal to the leg, so that the GRF passes through the COM. (c) Virtual leg from the toe to the COM, with length rv and angle θv
Parameter
Symbol
Value
Body mass
m
80 kg
Pitch inertia about the COM
J
5 kg m 2
COM–hip distance
dh
0.1 m
Leg rest length
r0
1 m
Leg stiffness
k
20 kN/m
Gravitational acceleration
g
9.8 m/s 2
Table 1: Physical parameters of the TSLIP model
Figure 2: Phases and events of a TSLIP stride: apex, touchdown, bottom, liftoff and next apex
Figure 3: One stride of the passive CPP runner from the upright fixed point ( αtd=109.8∘ ) with initial pitch θ0∈{0∘,5∘,−5∘} and zero pitch rate. The pitch stays constant (not shown). Forward lean injects energy and backward lean removes it. The shaded interval marks stance
Figure 4: Periodic gaits of the passive CPP runner. (a) Gaits in the (z∗,x˙∗) plane, colored by the touchdown angle; each touchdown angle yields a curve. (b) Transverse multiplier λ⊥ on a logarithmic color scale. Black squares have ∣λ⊥∣<1 . The three structural multipliers equal one in all cases. (c) Energy change during one stride when the gait is started with a constant lean of ±5∘
Figure 5: Largest closed-loop multiplier modulus with the stance PD law and the stride-level LQR, for all gaits of Fig. 4
Figure 6: Recovery from X0=[1.4m,2.91m/s,20∘,100∘/s]T under the PD–LQR controller, with unbounded hip torque and with ∣τ∣≤300 N m and μ=0.8 . (a)–(d) Apex states; dotted lines mark the nominal gait. (e) Peak hip torque and (f) peak friction ratio in each stance
Figure 7: Basins of attraction of the PD–LQR controller in the plane of initial apex height and speed. The star marks the nominal gait. Top row: unbounded hip torque; hatched regions converge only with a peak friction ratio above 0.8. Bottom row: hip torque and friction limits of Section IV.3 . Columns: upright start, initial pitch 20∘ , initial pitch rate 100∘ /s
Figure 8: Physical leg (r,α) , virtual leg (rv,θv) and the angle γ between them. Under CPP and the embedding of Section V , the COM moves as a point mass on a virtual leg with a prescribed radial force law
Figure 9: Decomposition of the apex return map into descent, stance and ascent maps with coordinate changes between Cartesian and virtual-leg coordinates
Quantity
Range
Grid
za
[1.1,1.5] m
(10)
x˙a
[0,8] m/s
(10)
θa
{−10∘,0∘,10∘}
(3)
θ˙a
set A: 0 ; set B: {7.5,15}∘ /s
(1); (2)
αtd
[90.5∘,124∘]
(19)
kv ( ω0 )
[8,36] kN/m ( [10.0,21.2] rad/s)
(10)
Table 2: Apex states and parameters used to evaluate the approximate maps. Grid sizes are given in parentheses. Set B uses five values each for za , x˙a and kv
NC
GC-M
GC-T
GC-L
SLIP
Set A ( θ˙a=0 )
plo
1.82 ± 1.64
1.28 ± 1.36
2.78 ± 2.35
0.70 ± 0.79
1.36 ± 1.34
p˙lo
7.16 ± 5.35
2.73 ± 2.05
2.63 ± 2.26
1.56 ± 1.83
1.75 ± 2.04
pa
4.14 ± 3.90
1.27 ± 1.18
2.47 ± 2.49
0.91 ± 1.22
1.69 ± 1.57
za
1.40 ± 1.89
0.70 ± 1.17
1.31 ± 1.57
0.30 ± 0.55
1.20 ± 1.48
Ea
1.86 ± 1.31
1.05 ± 0.83
0.45 ± 0.78
0.27 ± 0.22
0.01 ± 0.28
Table 3: Percentage prediction errors (mean ± standard deviation) of the approximate maps on admissible strides. NC, GC-M, GC-T: first-order expansion without correction, with mean-momentum and with total-momentum correction. GC-L: exact sweep with the linear-moment correction ( 34 ). SLIP: SLIP template with the approximation of [ 27 ] and energy correction. Strides for which a map predicts no flight phase are excluded for that map (set A: 0, 19, 68, 25 and 56 strides for the five maps; set B: 0, 8, 10, 6 and 4)
Figure 10: Mean prediction errors for set A; whiskers show one standard deviation. The bars of each quantity are ordered as in the legend
Figure 11: Mean prediction errors of (a) the liftoff velocity and (b) the apex position versus the stride asymmetry θtdv+θlov for set A (bins of 4∘ with at least 20 strides)
Figure 12: Mean prediction errors of (a) the liftoff velocity and (b) the apex position versus the natural frequency ω0 of the template for set A
Figure 13: Natural gaits of the passive embedded runner. (a) Apex speed versus apex height for three virtual stiffnesses at upright posture. (b) Touchdown angle versus apex height for five postures ( kv=20 kN/m). (c) Transverse multiplier; the second multiplier equals one at all gaits. Markers in (a)–(c): numerical gaits; lines in (a) and (b): closed form ( 39 ); lines in (c) join the numerical values. (d) Energy change per stride (hollow markers) and 100∥P(X)−X∥ (dots) along the symmetric families of P~ at αtd∈{110∘,113∘,116∘} , versus the touchdown force Ntd ; lines show Ntd2/(2kv) . The residual combines components in m and m/s
Figure 14: Deadbeat stride control of the embedded runner ( kv=20 kN/m). (a)–(c) Apex height, speed and pitch with targets (gray); (d) touchdown angle; (e) thrust; (f) actuator force ur at touchdown. Vertical dash-dotted lines: pitch-rate kick (stride 40) and push (stride 45)
Figure 15: Largest relative apex error in 400 random trials: median (bold lines) and 95th percentile (faint lines)
LIPM is everywhere in legged-locomotion control, but almost always as a modeling choice rather than as something the controller's cost actually prefers. This note tries to make that link more explicit. Working from a small centroidal OCP that penalizes the rate of angular momentum, we look at what its optimum tends to look like. Three things come out. With full-rank stance, the optimum drifts toward a pendular force pattern at a rate determined by the SVD of the moment Jacobian; the constant is set by foot-span geometry and matches the experiments to within 16%. With N=2 stance, as in trot, the friction cone introduces a lower bound on ∥H˙G∥ that no amount of weight tuning fixes; we also see a non-smooth feasibility kink at a critical horizontal acceleration that we can write in closed form. Adding a task term that asks for a nonzero H˙G moves the optimum off the pendular set in a predictable way. None of this is far from the classical ZMP/DCM picture. We test these claims on a point-mass quadruped and on the Unitree Go1 in MuJoCo (open-loop QP and a torque-level closed-loop controller), and we note where the asymptotic story stops being a good description of what the closed loop actually does.
Lingxue Lyu, Zihui Liu
School of Engineering and Applied Science, University of Pennsylvania, Philadelphia, PA 19104, USA · Department of Aeronautics & Astronautics, Stanford University, Stanford, CA 94305, USA
We present a novel stepping strategy for pitch unlocked planar monopeds where the reaction torques from stabilizing pitch with a conventional PD + feedfoward controller are recruited to counteract energetic losses from damping. By moving the location of the mass center, our controller increases the pitch stabilization torque, thereby adding energy to the gait. A new stepping policy adjusts the distribution of energy between the radial and angular degrees of freedom to counteract dissipative losses and achieve a user specified balance between steady state fore-aft speed and apex height. Hybrid averaging analysis yields closed form expressions for the fixed points and eigenvalues of the resulting gait, lending insight into the interplay between the physical and control parameters' influence on performance. Simulation studies on a generic 5 link biped and a careful model of the Penn Jerboa reveal a useful correspondence to these analytical predictions. Physical experiments on the Penn Jerboa exhibit stable locomotion with speeds ranging from 1.02 m/s to 1.77 m/s (5.10 leg lengths/s to 8.85 leg lengths/s) in a manner effectively approximated by the mathematical analysis.
Shane Rozen-Levy, Griffon McMahon, Daniel Koditschek
GRASP Lab, Department of Mechanical Engineering and Applied Mechanics, University of Pennsylvania, Philadelphia, PA, USA · GRASP Lab, Department of Electrical and Systems Engineering, University of Pennsylvania, Philadelphia, PA, USA
Despite decades of research, the dynamic mechanisms governing human locomotion remain incompletely understood. Current motor control models only partially capture these complex dynamics, a theoretical gap that likely limits the clinical efficacy of modern lower-limb rehabilitation therapies. Recently, it has been proposed that human locomotion is planned in task-space by explicitly exploiting the gravitational pull acting on the Center of Mass (CoM). Building upon this concept, this paper presents a novel deterministic task-space planner for straight, constant-speed locomotion. Our model formulates the CoM trajectory in the transverse plane as a harmonic oscillator moving at a constant forward velocity, while demonstrating that the vertical CoM trajectory is actively regulated through localized ankle strategies. We derive closed-form equations that successfully generate human-like trajectories for both the CoM and the swing foot. The model's predictive accuracy is empirically validated against the kinematic behavior of 12 healthy subjects walking at three self-selected speeds. Furthermore, the proposed planner is mathematically compatible with an extended formulation of the Passive Motion Paradigm (λ0-PMP), establishing a comprehensive theoretical foundation for decentralized, hierarchical architectures of semi-autonomous controllers in both human motor control and assistive robotics.
Carlo Tiseo, Ming Jeat Foo, Kalyana C Veluvolu +2
School of Engineering and Informatics, University of Sussex, Sussex House, Falmer, Brighton, BN1 9RH, England, United Kingdom · Rehabilitation Research Institute of Singapore, Nanyang Technological University, 50 Nanyang Avenue N3-01a-01, 639798, Singapore · School of Mechanical & Aerospace Engineering, Nanyang Technological University, 50 Nanyang Avenue N3-01a-01, 639798, Singapore +3