Source-linked AI summary
Ascento: A Two-Wheeled Jumping Robot
Victor Klemm, Alessandro Morra, Ciro Salzmann, Florian Tschopp, Karen Bodie, Lionel Gulich, Nicola Küng, Dominik Mannhart, Corentin Pfister, Marcus Vierneisel, Florian Weber, Robin Deuber, Roland Siegwart
TL;DR
Indoor ground robots need both fast movement on flat terrain and agile obstacle negotiation, capabilities that existing systems rarely combine. This paper presents Ascento, a compact two-wheeled jumping robot with dedicated mechanical and model-based control designs. Prototype experiments demonstrated stabilized driving, jumping onto a 10 cm step, and recovery from falls into various positions.
Problem
Indoor ground robots that combine fast flat-ground maneuvering with dynamic obstacle overcoming remain rare.
Method
The paper develops Ascento, a compact two-wheeled balancing robot using topology-optimized mechanical parts and dedicated controllers for stabilization, jumping, and fall recovery.
Results
Prototype experiments demonstrated stabilized driving, jumping onto a 10 cm step, and recovery from downfalls into various positions.
Takeaways & Limitations
Ascento provides a validated platform for fast flat-surface navigation, jumping obstacle traversal, and fall recovery in indoor settings.
Abstract
from arXiv · showhide
Applications of mobile ground robots demand high speed and agility while navigating in complex indoor environments. These present an ongoing challenge in mobile robotics. A system with these specifications would be of great use for a wide range of indoor inspection tasks. This paper introduces Ascento, a compact wheeled bipedal robot that is able to move quickly on flat terrain, and to overcome obstacles by jumping. The mechanical design and overall architecture of the system is presented, as well as the development of various controllers for different scenarios. A series of experiments with the final prototype system validate these behaviors in realistic scenarios.
I. INTRODUCTION
Ascento targets the gap between fast wheeled motion on flat ground and agile legged obstacle negotiation in compact indoor environments. The paper presents a two-wheeled jumping robot, its control software, and real-world prototype validation.
- Ground robots typically trade off fast, efficient flat-ground movement against effective handling of rough terrain and obstacles.
- Ascento combines wheels for efficient flat-ground travel with legs for overcoming obstacles by jumping.
- The robot uses a compact form factor for indoor spaces, with topology-optimized, 3D-printed structural components.
- Its control architecture uses LQR stabilization and sequential feed-forward control with feedback tracking for jumping and fall recovery.
- Real-world experiments with the prototype demonstrated stabilized driving, jumping, and fall recovery.
II. SYSTEM DESCRIPTION
Ascento’s mechanical design uses independently actuated wheeled legs and optimized linkage geometry to support compact height adjustment, stabilization, and jumping. Topology optimization and 3D printing reduce mass while maintaining structural strength.
- Each independently actuated leg can extend or retract, adjusting the robot’s total height between 31 cm and 66 cm.
- A three-bar linkage approximates linear wheel motion perpendicular to the ground and helps decouple stabilizing and jumping control.
- Topology-optimization-inspired leg geometries were designed to reduce mass and increase system strength.
- All structural parts are 3D-printed from PA12 using selective laser sintering, enabling fabrication of complex shapes and rapid prototyping iterations.
- Torsion springs counteract the robot’s weight, reduce hip-motor control effort, and increase efficiency and jumping height.
B. Hardware
The hardware combines high-torque actuators, compact direct-drive wheel assemblies, and sensing and software components for balancing, driving, jumping, and recovery. The complete system weighs 10.4 kg and operates for approximately 1.5 hours.
- Hardware: ANYdrive series-elastic actuators provide hip-motor peak torques of up to 40 Nm for jumping and spring counteraction.
- Hardware: Custom frameless hub-motor wheel assemblies provide compact, near-zero-backlash motion and high wheel torque for stabilization.
- Hardware: 10.4 kg system weight and approximately 1.5 h operation time characterize the complete robot.
- Software: A Kalman filter combines IMU and motor-encoder data with model knowledge to estimate the robot’s state for control.
- Software: The controller block includes stabilizing, jumping, fall-recovery, and high-level position controllers, with driving and jumping controlled independently.
1) Notation convention:
The paper defines state and generalized-coordinate variables for modeling Ascento, then derives its dynamics from a simplified rigid-body model with explicit assumptions and measured parameters.
- Notation convention: θ denotes forward tilt, v planar linear velocity, and ω normal angular velocity; β and h describe leg configuration.The model also uses x, y, γ, and s for planar odometry.
- Notation convention: The generalized-coordinate vector q = [θ s γ]⊺ models the system, while x = [θ ˙θ v ω]⊺ provides stabilization feedback.
- Model assumptions: The model assumes rigid bodies, no-slip wheel contact, frictionless perfect joints, neglected motor dynamics, and no system delay.Fixed leg geometry limits applicability to a specific leg configuration, addressed later through interpolated control.
- Rigid-body model: Fixed leg geometry reduces the robot to a two-wheeled inverted pendulum comprising two wheels and a substitute-mass pendulum body.
- Rigid-body model: Kinematics and body energies yield a Lagrangian model whose equations of motion use mass matrix M and forcing term f.Parameters are obtained from component mass measurements and CAD models under constant-density assumptions.
IV. CONTROL
Because a two-wheeled robot is inherently unstable, Ascento requires dedicated control strategies for standing, driving, jumping, and recovery after falling.
- IV. CONTROL: A two-wheeled robot is inherently unstable and therefore needs dedicated control during standing still and driving.
- IV. CONTROL: Jumping requires a dedicated control strategy in addition to stabilization and driving control.
- IV. CONTROL: Entering operational mode or recovering from a fall requires a specific control maneuver.
A. Stabilizing Control
Ascento stabilizes its inherently unstable body with an interpolated LQR controller built from linearized models across leg heights and driven by wheel torques.
- A. Stabilizing Control: LQR stabilization regulates the linearized system at minimal cost while targeting reliability, robustness, and compact disturbance recovery.
- A. Stabilizing Control: The infinite-horizon LQR solution is obtained from the discrete-time algebraic Riccati equation, which determines the optimal feedback gain K.Q and R encode state importance and overall control aggressiveness.
- A. Stabilizing Control: The LQR state vector x = [θ ˙θ v ω]⊺ omits spatial position and orientation, delegating position tracking to a velocity-command controller.Operator setpoints are added to v or ω to command target velocities.
- A. Stabilizing Control: Ten equally spaced leg-height linearizations produce ten fourth-order state-space models and interpolated feedback gain matrices.Interpolation addresses the fixed-leg-geometry restriction in the simplified model.
- A. Stabilizing Control: The controller uses left and right wheel torque as inputs, with estimated leg height selecting an interpolated gain matrix.
B. Jump Control
The jump controller takes over from drive control and executes a predefined, phase-based feed-forward sequence, while stabilization remains active during leg retraction.
- B. Jump Control: Activating the jump controller overrides the drive controllers and gives full control to a predefined jump sequence.
- B. Jump Control: The heuristic feed-forward jump controller uses discrete successive phases inspired by human jumping motion.Each phase applies a different control strategy.
- B. Jump Control: During leg retraction, a hip-motor trajectory controller operates while the stabilizing controller remains active.
2) Trigger Jump:
The jump controller coordinates leg extraction, spring-damper behavior, ground-contact detection, and user-selected jump parameters for different obstacle scenarios.
- After nominal stability, the robot accelerates forward and triggers leg extraction at a predefined distance from the step using ToF sensors.
- The hip motors synchronously extract the legs under PID control until ground contact is lost.
- A virtual spring-damper behavior at the hips supports jumping over high obstacles or preventing wheel contact with a stair edge.
- Ground contact is detected from hip-joint torque thresholds, after which stabilization resumes and hip compliance dissipates landing energy.
- Users can set jump height and forward velocity through a GUI for jumping on the spot, while driving, or onto a step.
1) Retract Legs:
The stand-up and fall-direction procedures use prescribed hip trajectories and wheel torques to move the robot from resting configurations toward upright balance.
- The robot retracts its legs by following a specific hip-motor trajectory.
- For the laying position, controlled leg extraction rotates the robot around its knee into a sitting position.Extraction speed must balance maintaining ground contact against generating enough rotation to tip the robot over.
- Constant backward or forward wheel torque supplies rotational energy from sitting or planking positions until the robot reaches zero tilt angle.
- The robot enters a resting position by disabling stabilization and applying a small wheel torque to control the fall direction.
V. EXPERIMENTS
Experiments evaluate simulation fidelity and prototype capabilities in stabilization, jumping, and fall recovery, including jumping onto a 10 cm step.
- Simulation used Gazebo with CAD-derived mass and inertia values, and similar simulated and experimental behavior supported the model’s validity.
- Prototype experiments tested stabilizing performance, jumping, and fall recovery, with robot dimensions and technical specifications also reported.
- The technical specifications are presented in Table II, while Figure 8 shows the robot’s main dimensions.
- The robot remains upright under large external disturbances and recovers from impulsive, persistent, and longer-lasting disturbances.
- 10 cm: the prototype jumps onto small steps, requiring at least 90 cm before and after the step for acceleration and safe landing.
3) Fall Recovery:
Fall recovery combines controlled stand-up behavior with additional experimental capabilities, while the conclusion identifies jumping, disturbance recovery, and future terrain adaptation as key directions.
- 3) Fall Recovery:: Less than 2 m: standing up from all recoverable resting positions requires this amount of space.
- 3) Fall Recovery:: The prototype’s fall-recovery demonstrations are part of a broader set of experiments validating its capabilities.
- 3) Fall Recovery:: Figure 11 shows the robot standing up by first reaching a sitting position and then the standing position.
- Experimental Features: Exploration, collision-free path following, force-compensating lean, and possible one-leg balancing or MPC remain experimental features validated only in simulation.
- Conclusion: The conclusion reports fast flat-surface navigation, obstacle jumping, an impact-resistant lightweight design, model-based LQR control, and prototype demonstrations of step jumping and fall recovery.