Compliant Mechanisms (CMs), along with Soft Robotic devices
formed therewith, may be defined as engineering systems
achieving force and motions transmission via the deflection
of flexible members. CMs have increasingly gained a strong
foothold in the scientific arena owing to their hinge-less
nature, shock resistance, potential single-piece
manufacturability, safety in human-machine interaction,
minimal maintenance requirements, and adaptability to work
in unstructured environments. In parallel, current advances
in the production of inherently compliant sensory-motor
apparatus, as well as progresses in the development of
robust control methods, are paving the way to practical CM
adoption in a large variety of engineering fields, here
including healthcare, manufacturing,
inspection/maintenance, and agrifood. In light of these
considerations, the objective of the present talk is to
critically discuss, on the basis of our experience,
engineering methods and related Computer-Aided Engineering
(CAE) tools allowing for the (automatic) computation of
accurate CM models and for the shape optimization of
complex-shaped CMs comprising out-of-plane displacements,
distributed compliance, and non-linear materials. Within
this scenario, a strong integration between different
design tools is desirable to improve the work of robotics
engineers, reducing the number of errors and speeding up
the design process. We hence present a strong integration
between 3D computer-aided design models and multidomain
simulations applied to the design of compliant,
tendon-driven, robotic hands/grippers. The integrated
design framework is employed to optimize: i) a
four-fingered reconfigurable gripper comprising highly
deformable components; ii) a compliant anthropomorphic
hand/wrist system for possible usage in humanoid robotics.

This paper presents a quasi-static analytical model for
normal force distribution in a skid-steering mobile robot
operating on uneven terrain. The model accounts for terrain
inclination through pitch and roll angles, enabling the
analysis of load transfer between wheels. In addition,
interaction forces arising from wire connections are
incorporated, extending the formulation to physically
connected multi-robot systems. Unlike conventional models
that consider isolated robots, the proposed approach
captures internal coupling effects introduced by flexible
links. The model is based on equilibrium equations and
provides explicit expressions for normal forces as
functions of terrain and system parameters. The formulation
is particularly suitable for systems such as RoboShepherd
and provides a foundation for further dynamic analysis and
control design in off-road multi-robot applications.

Abstract—Motion control is a fundamental requirement for
any mobile robot operating in a time-critical environment
such
as a robotics competition, where every trajectory is a
sequence
of moves spanning a wide range of amplitudes that the chosen
controller must handle reliably. A fixed-gain PID is the
stan-
dard starting point for differential drive robot (DDR)
motion
control, yet its constant gains limit adaptability, which
is a
property that this study examines across move amplitudes,
with
payload and surface variation left for future work.
Conversely,
more elaborate control architectures sometimes add
complexity
without delivering a measurable benefit. This paper presents
an experimental comparison of five motion control
architectures
implemented on the same physical differential-drive robot
built
for the Eurobot 2026 competition: a decoupled dual-channel
PID
used as a baseline, a gain-scheduled PID, a pure feedforward
controller based on a trapezoidal velocity profile, a
feedforward
controller combined with PID feedback, and a fuzzy PID.
All architectures are evaluated under identical conditions
on
two elementary motion primitives - straight-line
point-to-point
translation and rotation in place - at multiple magnitudes.
Performance is assessed using root mean square error (RMSE),
integral of absolute error (IAE), settling time, overshoot,
and
maximum control effort. The feedforward + PID achieves the
best overall performance, and the results provide
evidence-based
guidance for architecture selection in competitive and
educational
differential-drive applications.

Maintaining balance of legged robots in two-legged stance
remains challenging due to underactuation, nonlinear
dynamics, and unilateral ground contact. This paper
presents balance control of a bipedal robot using a linear
quadratic regulator (LQR) based on two simplified models:
the acrobot and flywheel inverted pendulum. The developed
control approach is implemented on the Solo12 quadrupedal
robot, which is required to balance on only two legs.
Simulations are carried out in Julia and in the MuJoCo
physics simulator. The control law explicitly enforces
contact constraints, ensuring positive vertical ground
reaction force and maintaining horizontal force within
friction limits. Results show that the acrobot-based LQR
stabilizes the system for small initial angle deviations
and brief external perturbations, but fails for larger
offsets. The flywheel-based LQR exhibits a wider stability
region in Julia but does not transfer well to MuJoCo due to
discrepancies between the flywheel model and the actual
dynamics of the Solo12 robot.

In this paper, we present AlignIt,
a self-supervised visual servoing system for robot arm
alignment that requires no
CAD model, markers, tags, or manually labeled training
data. An autonomous spiral trajectory (50 waypoints,
10 episodes, 500 data sets in total) is executed above a
single reference
pose defined by the operator, thereby generating a densely
labeled set
of RGB images paired with relative pose errors in SE(3)
space,
encoded as 9D vectors using the continuous 6D rotation
representation. An EfficientNet-B0-based regression network
maps a single RGB image to the corresponding 9D pose-error
vector.
During deployment, the network's predictions are decoded
back into SE(3)
transformations and applied iteratively as position servo
control
commands on the UFactory Lite 6 robot arm, with the system
converging
when the positional error falls below 2mm and the
orientation error below
4 degrees for 10 consecutive steps. In this implementation
we used an
Intel RealSense D435i camera, without prior knowledge of the
object CAD model or scene geometry.

The pull-up is a fundamental closed kinematic chain
exercise widely used in strength training, yet its
biomechanics remain under-investigated compared with
lower-extremity tasks. This paper presents a complete,
from-scratch workflow for the frontal-plane inverse-dynamic
analysis of pull-ups. Five young healthy participants
performed pull-ups in a laboratory setting while
retroreflective markers were tracked by an optical
motion-capture system. A planar rigid-body model of the
body and upper limbs was established, body-segment inertial
parameters were assigned from published anthropometric
data, and inverse kinematics, followed by inverse dynamics,
were implemented from first principles. Joint angles, net
joint moments, and joint powers at the shoulder and elbow
are reported and discussed.
Results are consistent with the expected bilateral loading
pattern and reveal the dominant role of the shoulder
adduction moment during the concentric phase.
Across repetitions, decreasing mean propulsive velocity
(MPV) together with changes in complementary fatigue
metrics indicate progressive alterations in coordination
and control. The dataset will be made available to the
community.

Biomechanical simulation and analysis of human motion
requires both anatomically plausible body models and
reliable data-processing pipelines. This paper presents two
open-source Python tools for MuJoCo: one for procedural
generation of anthropometrically consistent human models in
the MuJoCo XML format from height, weight, and biological
sex, and one for motion-capture processing with inverse
kinematics, inverse dynamics, and ground reaction force
estimation. By leveraging MuJoCo's contact-model-based
inverse dynamics, the pipeline requires only kinematic data
of anatomical landmark points as input for biomechanical
reconstruction. Validation on a countermovement vertical
jump (CMJ) dataset shows that joint-angle and joint-moment
trajectories remain consistent with reference values
reported in the literature, while estimated ground reaction
force profiles show moderate agreement with measured forces
and leave room for future improvement through automatic
parameter calibration.

Frictional contact simulation is a fundamental challenge in
computer graphics and robotics, requiring a delicate
balance between physical accuracy, computational
performance, and numerical stability. This paper begins
with a foundational analysis of the challenges inherent in
contact modeling, detailing the trade-offs that confront
traditional simulation methods. Building upon this
analysis, we present our implementation of the Staggered
Projections (SP) algorithm, a velocity-level formulation
for resolving frictional contact in multibody systems. We
explore the core principles of the method, which decomposes
the complex, coupled contact problem into a sequence of
more manageable projection operations. Our evaluation
demonstrates the algorithm's capacity to robustly handle
challenging scenarios, such as large-scale stacking and
friction-dependent structures. However, its most
significant limitation lies in the high computational
complexity in complex environments.

U radu se razmatra problem optimizacije proizvodnog
rasporeda u uslovima velike raznovrsnosti i malih serija
(eng. High-Mix, Low-Volume, HMLV) sa ciljem smanjenja
troškova električne energije. Posmatrani sistem obuhvata
kolaborativnog robota koji učestvuje u procesu montaže dve
vrste fiskalnih kasa, pri čemu različite operacije robota
imaju različitu energetsku potrošnju. Optimizacija je
sprovedena u dva koraka: određivanje optimalnog rasporeda
delova po kutijama u cilju minimizacije potrošnje energije,
i optimizacija vremenskog rasporeda proizvodnje u skladu sa
promenljivim cenama električne energije. Za rešavanje
problema korišćen je metaheuristički algoritam simuliranog
kaljenja, koji omogućava efikasno pretraživanje velikog
prostora mogućih rešenja. Rezultati simulacija pokazuju da
predloženi pristup može ostvariti značajne uštede troškova
električne energije, naročito u uslovima manjih proizvodnih
serija, karakterističnih za HMLV okruženje. Dobijeni
rezultati ukazuju na to da inteligentno planiranje
proizvodnje može unaprediti energetsku efikasnost bez
dodatnih ulaganja u opremu, što je od posebnog značaja za
mala i srednja preduzeća.