We propose a hierarchical control framework to address severe dynamic instabilities and navigational drift that arise when a two-wheeled inverted pendulum (TWIP) robot attempts asymmetric object manipulation. While two-wheeled platforms are highly manoeuvrable, their constant balancing adjustments make onboard odometry highly unreliable for precise navigation. Furthermore, the addition of a side-mounted robotic arm introduces unactuated lateral roll moments when a payload is lifted, a challenge heavily compounded on uneven terrain. To solve these coupled problems, our architecture divides the workload. An offboard vision system tracks overhead ArUco markers to provide high-latency global waypoint navigation, bypassing odometry drift. Simultaneously, a low-latency onboard control loop rejects active physical disturbances using inertial and encoder data. In our physical experiments, this dual-loop approach enabled the custom-built robot to navigate accurately, reject transient impacts from speed bumps, adapt to a dynamic seesaw ramp, and carry a payload securely without falling over its narrow wheelbase.
Figures & tables
Fig. 1: System architecture and workspace overview. The offboard Central Control Unit processes a global visual feed to track ArUco fiducial markers. Spatial coordinates and trajectory waypoints are transmitted via Bluetooth to the onboard ATmega328P microcontroller, which governs the low-latency dynamic balancing and asymmetric manipulation tasks.
Fig. 2: Hardware architecture and mechanical assembly of the TWIP platform. (a) An exploded 3D CAD model detailing the structural components, custom off-axis manipulator, and placement of the overhead ArUco fiducial. (b) The physical control stack (top view, lid removed) housing the ATmega328P microcontroller, L298N motor driver, and communication modules, alongside the drive assembly (bottom view) featuring the coaxial N20 motors, magnetic encoders, and the MPU6050 IMU.
Fig. 3: Integrated kinematics and dynamic response of the off-axis manipulator. (a)-(d) The kinematic sequence of payload acquisition: approach, engagement, securing, and lifting. (e) The corresponding difference in motor control effort during the asymmetric payload lift. The shaded servo actuation phase correlates with step (d), illustrating the permanent steady-state PWM increase required on the loaded side to counteract the induced lateral roll moment.
Fig. 4: (a) The side CAD view of the asymmetric platform. (b) Free-body diagram (FBD) of the wheeled inverted pendulum, including the pitch state ( θ ) and the longitudinal translation ( x ) of the centre of mass.
Fig. 5: Front-facing schematic of the asymmetric lateral roll disturbance. The off-axis payload mass ( mp ) at distance dy generates an unactuated torque ( τroll ) that must be countered by the 13.0 cm wheelbase ( W ).
Fig. 6: The system architecture demonstrates a dual-loop control framework. Visual tracking with ArUco fiducials is managed by the offboard Central Control Unit (CCU), which enables the onboard ATmega328P to focus on Linear Quadratic Regulator (LQR) balancing, motor Pulse Width Modulation (PWM) updates, and servo actuation in response to Bluetooth commands.
Fig. 7: Overhead visual navigation sequence and 2D spatial path tracking. (a)-(d) The CCU tracks ArUco fiducials calculating distance and heading. (e)-(f) The robot executes payload acquisition. (g) The 2D spatial path tracking overlays the physical path against the theoretical trajectory.
Fig. 8: The experimental validation of transient impact rejection. (a)-(b) show the dimensions and render of the 5 mm speed bump. (c)-(e) depict the empirical sequence of the robot going over the obstacle. (f) is a time-domain representation of the pitch response which shows the rapid damping of the LQR controller.
Fig. 9: Experimental validation of dynamic pitch variation over a pivoting ramp. (a)-(b) Schematic and render of the seesaw mechanism. (c)-(f) Sequential execution of the robot transporting an asymmetric payload. (g) Dynamic pitch disturbance rejection showing the robot successfully adjusting its local pitch reference.
Department of Electronics and Communication Engineering Krishna Institute of Engineering & Technology (KIET), Ghaziabad, Delhi-NCR, Uttar Pradesh, India.
Department of Electronic and Computer Engineering, The Hong Kong University of Science and Technology, Hong Kong, China. · X-lab Department, Agibot Innovation Company, Shanghai, China.