IS-431: Robot Mechanism

5.2 Robot Mechanism

The robot mechanism encompasses all physical hardware, including mechanical structures, actuators, sensors, and supporting electrical and firmware elements, that together enable BoxBunny to operate as a realistic and interactive boxing training partner rather than a static target. At the system level, five principal functions are served by the mechanism: stable physical support capable of resisting punching-induced overturning moments; motorised yaw rotation for re-angling the upper body to present changing attack and defence lines; motorised vertical adjustment to match the robot's striking targets to users of different statures; impact-absorbing contact surfaces with embedded sensing for training analytics; and a high-velocity 2-DOF arm actuation mechanism capable of delivering anatomically correct jabs, hooks, and uppercuts.

It is noted that full footwork movement technically requires both linear forward-backward translation and rotational yaw motion. Mechanical development was phased to prioritise high-quality execution within the available timeframe: Phase 1 addressed static stability and the height adjustment subsystem, while Phase 2 focused on the rotation subsystem. Linear movement has therefore been scoped out of the current mechanical prototype.

These functions were decomposed from the broader project goals established during the problem clarification and design methodology phases. A key integration principle across all subsystems is that the mechanism was designed from system-level behaviour downward, rather than by combining isolated components. The robot was first required to remain upright, safe, and spatially compact within a boxing environment. Only after these high-level requirements were defined were they decomposed into subsystem-level design criteria following the left side of the V-model, as described in the System Design Narrative below.

Design Requirements

This subsystem addresses DO-2 (Intelligent Sparring System), DO-4 (Adaptive Fight Intelligence), and DO-5 (Modular Boxing Platform). See Section 5 for the full Design Objectives reference.

The following system-level requirements govern the entire robot mechanism. Each subsystem page describes how these system requirements cascade into subsystem-specific design criteria.

ID	Subsystem	Requirement	Rationale
RM-1	Base	Remain upright under worst-credible punching loads with FoS ≥ 1.5	User safety; conservative margin for uncertainty in punch intensity and direction
RM-2	Base	Compact footprint that preserves user footwork space	Boxing gyms are space-constrained; base must not obstruct pivots or stance changes
RM-3	Base	Portable: transportable by 1 person between venues	Demo and gym deployment flexibility
RM-4	Rotation	Yaw rotation for re-angling with target angular velocity ~150°/s	Replicate coach pivoting behaviour for realistic defensive and counter drills
RM-5	Height Adjustment	Provide ≥ 400 mm vertical stroke for height adjustment	Accommodate user height range for anatomically correct target alignment
RM-6	Padding	Absorb repeated strikes without damage; provide impact detection across all three target zones	Structural survivability and training analytics
RM-7	Arm Actuation	Deliver three distinct strike types: Jab (pitch), Hook (roll), and Uppercut (pitch + roll)	Strike variety drives the 2-DOF coaxial differential joint concept selection and supports DO-3 drill progression
RM-8	Arm Actuation	Execute a 90° arm sweep in ≤ 0.25 s	Training fidelity requires realistic strike velocity; target remains ≤ 0.25 s (system requirement). Best recorded result: 0.64 s (Partial — Damiao PID acceleration overhead)

System Design Narrative

Following the V-Model framework, system-level requirements RM-1 through RM-7 were established before any subsystem design was initiated. These requirements were then decomposed into subsystem-level engineering criteria, each driving a dedicated development effort across mechanical, electrical, and firmware domains. The diagram below illustrates the BoxBunny-specific V-model: design decomposition on the left, fabrication and integration at the base, and verification closure on the right. Horizontal dashed lines indicate the correspondence between each design phase and its verification counterpart.

_{Figure: BoxBunny Systems Engineering V-Model — design decomposition (left, blue), fabrication and integration (centre, amber), and verification closure (right, green). Dashed lines indicate requirement-to-verification correspondence at each level.}

Structural Foundation: Base and Rotation RM-1, RM-2, RM-4

The trapezoidal base geometry resolves the tension between RM-1 (stability under punching loads) and RM-2 (compact footwork clearance) through a single geometric decision: the plan form is narrower at the front and wider at the rear, simultaneously increasing the restoring moment and reducing intrusion into the boxer's working zone. The rigid mounted plate at the top of the base serves as the geometric datum for the entire mechanism stack above. At the base of the stack, the rotation subsystem achieves RM-4 via a non-geared slewing bearing paired with an off-axis timing-belt drive, with outboard cam-follower supports extending anti-tilt resistance without enlarging the bearing footprint; full derivation is provided in Sections 5.2.1 and 5.2.2.

Adaptive Height Targeting: Height Adjustment RM-5

The critical design insight is the separation of vertical lift from lateral structural resistance within a single assembly: the screw jack operates in axial compression only, while all punching-induced lateral loads are resolved by the telescopic column structure and Delrin wear-pad interfaces. This decoupling prevents screw binding and provides inherent fail-safe position retention through the self-locking geometry of the jack. Full design rationale and load analysis are documented in Section 5.2.3.

Impact Absorption and Sensing: Padding RM-6

The multi-layer padding architecture resolves the tension between energy absorption and sensing fidelity by separating these functions across distinct physical layers: polyethylene foam absorbs surface energy, anti-vibration isolation mounts protect the 3D-printed arm drivetrain, and MPU6050 IMUs are positioned at the interface layer where sufficient acceleration signal remains. Two firmware-level defects — an I²C bus hang caused by parasitic capacitance on long wires, and a Nyquist blind-spot in punch detection — were identified and resolved during integration. Full mechanical and electrical details are in Section 5.2.4.

Strike Delivery: Arm Actuation RM-7 & RM-8

The 2-DOF coaxial differential joint enables three strike types (jab, hook, uppercut) from a single compact pivot, with both actuators co-located at the pivot to minimise moment of inertia. Actuator platform selection iterated through servo and ODrive configurations before settling on the Damiao DM-J4310-2EC; a post-CDE Fair material revision replaced polymer shafts with a 6 mm stainless steel D-shaft to achieve sustained sparring capability. RM-8 is partially fulfilled: the arm's kinematic design is sufficient, but Damiao PID acceleration and deceleration ramps impose approximately 0.4 to 0.5 s of constant overhead per strike; the full analysis is in Section 5.2.5 Limitations.

System Integration

Three cross-cutting decisions unify all five subsystems: a dual-rail power architecture isolating the 24 V motor bus from the 12 V logic rail, preventing regenerative braking events from disrupting the control layer; a two-tier control hierarchy separating the deterministic 200 Hz firmware loop on the Teensy 4.0 from combat decision logic on the Jetson Orin NX under ROS 2; and a WiFi UDP link for the rotation axis, eliminating cable fatigue through the rotating base joint. The rigid mounted plate of the base closes the mechanical interface chain, propagating subsystem alignment upward through the entire stack.

Overall Subsystem Architecture

Mechanical Stack

The mechanism is organised as a vertical stack, ordered from the ground up. This ordering principle ensures that the heaviest and most load-critical components sit closest to the base, where they can best resist overturning moments:

_{Interactive 3D Model: BoxBunny Mechanical Stack}

Trapezoidal Base : welded steel platform providing stability and portability
Rotation Yaw Stage : low-profile rotation module at the stack base
Telescopic Height-Adjustment Column : motorised screw-jack column with structural guiding
Upper Body : torso, head, and 2-DOF coaxial striking arms with multi-layer padding

The rotational footwork axis is fully motorised for programmable, repeatable training sequences, allowing the robot to dynamically pivot and present changing attack lines. Meanwhile, the height adjustment uses a self-locking manual screw jack for safety and structural rigidity, ensuring the upper body remains stable under dynamic boxing loads.

Electrical Architecture

The system is unified under a dual-rail power architecture that electrically isolates the high-current 24 V motor bus from the 12 V logic rail serving the compute and control subsystems. Each mechanical axis is served by a dedicated motor, driver, and controller matched to its kinematic and safety requirements: the arm actuation subsystem employs integrated brushless servo actuators communicating over a shared CAN bus; the height adjustment axis uses a brushed DC gear motor driven through a dual H-bridge driver with regenerative braking protection; and the base rotation axis is governed by a CAN Modbus driver commanded wirelessly from the main compute module. Impact detection is provided by inertial measurement units embedded within the padding structure, each communicating over independent I2C buses to the central microcontroller. Full motor specifications, driver configurations, bus addressing, and power budget derivations are documented in the respective subsystem pages.

Data Architecture

The system implements a two-tier hierarchical control architecture. The upper tier resides on an NVIDIA Jetson Orin NX compute module running ROS 2 Humble, which hosts the graphical control interface, executes the combat finite-state-machine, sequences strike libraries, and processes impact detection events. The lower tier resides on a Teensy 4.0 microcontroller executing a deterministic unified control loop that manages CAN command transmission to the arm actuators, PWM output to the height driver, I2C polling of all inertial sensors, and publication of a consolidated feedback payload to the Jetson via a micro-ROS USB bridge. A hardware-level current watchdog running independently of the GUI provides a last-resort safety layer by immediately disabling all arm actuators if phase current on any axis exceeds its rated threshold. The base rotation axis is governed by a dedicated controller executing on a secondary WiFi-enabled microcontroller, receiving angle references from the Jetson over a UDP link. The height axis operates under open-loop timed commands, with passive position retention provided by the self-locking geometry of the screw-jack mechanism. Impact events are classified on the Jetson by computing the acceleration vector magnitude from inertial sensor readings and publishing detected strikes as timestamped ROS 2 messages. Control loop rates, ROS 2 topic definitions, bus utilisation strategies, and watchdog parameters are detailed in the individual subsystem pages.

Individual subsystem pages detail the specific electrical interfaces, power budgets, bus configurations, and control parameter derivations for each axis.

Verification Plan

Following the V-Model framework introduced in Section 3.2, the verification plan defines how each system-level requirement (RM-1 through RM-8) will be verified at the subsystem and system levels. Verification plans are established during the decomposition phase (left side of the V) and executed during integration (right side).

Subsystem Verification Matrix

Each system-level requirement cascades into a specific subsystem test. The table below defines the mapping, test method, and acceptance criterion for each:

Req	Subsystem	Test	Acceptance Criterion	Status
RM-1	Base	Tipping test: apply worst-credible punch load at max height extension	No tipping; restoring moment / overturning moment ≥ 1.5	Passed
RM-2	Base	Footwork clearance: user completes full stance cycle around robot	No foot contact with base during orthodox and southpaw stances	Passed
RM-3	Base	Transport test: 1-person tip-and-roll between two venues	Successful transport without disassembly; ≤ 5 min setup	Passed
RM-4	Rotation	Angular velocity measurement: 90° step command, measure time-to-target	Sustained angular velocity ≥ 150°/s	Pending
RM-5	Height Adj.	Full-stroke actuation: command min → max height	≥ 400 mm travel; ≤ 32 s full stroke	Pending
RM-6	Padding	60-punch impact detection test across 3 body zones (20 per zone)	≥ 95% true positive rate; monotonic force differentiation	Pending
RM-7	Arm Actuation	Strike type demonstration: all six strike variants (Left/Right Jab, Hook, Uppercut)	All six strikes delivered correctly in sustained sparring session	Pending
RM-8	Arm Actuation	Strike speed timing: command 90° arc, measure completion time	90° in ≤ 0.25 s (jab, hook, and uppercut) — Partial; best result 0.64 s	Partial — kinematic speed achieved; Damiao PID ramps add ~0.4–0.5 s overhead. See Section 5.2.5 Limitations.

System Integration Test Plan

Beyond individual subsystem verification, the robot mechanism must be tested as an integrated system — all five subsystems operating together under realistic training conditions (without Robot Intelligence). The system-level integration tests validate cross- subsystem interactions that cannot be caught by isolated subsystem tests:

Test	Description	Acceptance Criterion
Coordinated Motion	Height adjustment + yaw rotation simultaneously under operator control	No mechanical binding, smooth motion, no loss of position
Loaded Sparring	Strike delivery while user is actively punching the padding	Arm maintains strike accuracy; IMU correctly detects user punches during robot strikes
Power Budget	Measure total system current draw under worst-case concurrent actuator load	Total draw within PSU capacity with ≥ 20% headroom
Communication Reliability	Run CAN + I²C buses concurrently at full rate for 30 min continuous sparring	Zero bus hangs; ≤ 0.1% packet loss
Safety E-Stop	Trigger emergency stop during active strike motion	All motors decelerate to zero within 100 ms; no uncontrolled motion

Detailed test results are documented on the testing page:

Verification Results

Subsystem test results, system integration results, performance assessment

Subsystem Documentation

The robot mechanism is decomposed into five subsystems, ordered from the base of the mechanical stack upward. Each subsystem page follows a consistent structure: requirements recap (how system-level requirements cascade down), design rationale and evolution, and validation against subsystem criteria. Select a subsystem below to view its detailed documentation.