Back to Robot Mechanism

5.2.5 Arm Actuation

The arm actuation subsystem constitutes the primary strike-delivery mechanism of BoxBunny. Its function is to replicate the upper-body actions of a human training partner, executing targeted punches at anatomically derived striking zones and providing real-time impact detection for training analytics. This page presents an executive summary of the subsystem design, from requirements through to validated test results. Detailed derivations are linked in each subsection below.

Design Requirements

The arm actuation subsystem responds to system-level requirements RM-7 (strike types) and RM-8 (strike speed), both defined in the Robot Mechanism section (Section 5.2). The subsystem requirements below elaborate on how those system requirements are realised at the arm joint level.

System Requirements

ID Requirement Source
RM-7 Deliver three distinct strike types: Jab (pitch), Hook (roll), Uppercut (pitch + roll) Padwork rehabilitation literature; Robot Mechanism (Section 5.2)
RM-8 Execute a 90° arm sweep in ≤ 0.70 s; original 0.25 s target revised to account for Damiao PID acceleration overhead Robot Mechanism (Section 5.2); see Testing and Evaluation (Section 5.2.5.5) for PID overhead analysis

Subsystem Acceptance Criteria

The following acceptance criteria are derived from specific design decisions that introduce failure modes not captured at the system requirement level. ARM-AC-1 arises from ARM-SR-1 (90° motor sizing) — the choice of Damiao DM-J4310-2EC with internal PID ramps introduces a speed failure mode requiring firmware-measurable timestamp validation. ARM-AC-2 arises because RM-8 does not constrain motor thermal behaviour — sustained sparring introduces a thermal runaway failure mode requiring temperature monitoring (< 60°C). ARM-AC-3 arises from the dual-supply power architecture — the isolated 12 V logic rail must survive repeated emergency stops without PSU OVP trips, a failure mode not addressed by any RM code. ARM-AC-4 arises from the combat FSM design — multi-strike sequencing introduces a chain-execution latency failure mode not captured by single-strike RM-8. ARM-AC-5 arises from the current-limit safety system — peak motor current must stay below 2.0 A to protect 3D-printed PLA gears from thermally induced fatigue.

ID Acceptance Criterion Derives From Verification Test
ARM-AC-1 90° arm sweep ≤ 0.70 s (all six strike types at 30 rad/s) ARM-SR-1 / RM-8 — Damiao PID ramp speed failure mode Firmware timestamp delta (N=43 speed tests)
ARM-AC-2 5 min continuous sparring; peak motor temp < 60°C RM-8 / RM-7 — thermal failure mode under sustained high-duty-cycle operation Continuous sparring session + Damiao temperature telemetry
ARM-AC-3 No PSU trip across 5 consecutive emergency stops Dual-supply power architecture — regenerative braking OVP failure mode Visual confirmation + voltmeter monitoring
ARM-AC-4 3-strike chain executed in ≤ 5 s total Combat FSM design — multi-strike sequencing latency not captured by single-strike RM-8 Video timestamp + GUI console log
ARM-AC-5 Peak motor current < 2.0 A per motor (all strikes) PLA gear thermal protection — current-limit safety system failure mode CAN current telemetry (N=43 speed tests)

Table: Arm Actuation Subsystem Requirements and Acceptance Criteria

System Design Narrative

Following the Systems Engineering V-Model introduced in Section 3.2, RM-7, RM-8, and ARM-AC-1 through ARM-AC-5 were fixed before detailed motor selection, gear ratio sizing, or firmware architecture decisions were made.

BoxBunny Arm Actuation — Systems Engineering V-Model V-Model diagram for the BoxBunny boxing robot Arm Actuation subsystem. BoxBunny Arm Actuation — Systems Engineering V-Model DESIGN DECOMPOSITION VERIFICATION AND VALIDATION 01 System Requirements RM-7, RM-8 ARM-AC-1…5 defined 02 Design & Ideation 2-DOF coaxial differential Mass centralisation strategy 03 Mechanical & Electrical 3:1 helical gear / steel D-shaft Dual-supply 24 V / 12 V rail 04 Integration Build V5 arm assembly + calibration Strike library active 05 Firmware & Control 200 Hz CAN unified loop Kinematic decoupling verified 06 Speed & Safety Tests N=43 benchmark (all strikes) ARM-AC-1 / ARM-AC-5 → Pass 07 Validation RM-7, RM-8 ARM-AC-1 through ARM-AC-5 Design Decomposition Integration Build Verification & Validation Design arm V&V arm

Figure: Systems Engineering V-Model for the BoxBunny Arm Actuation subsystem, tracing RM-7, RM-8, and ARM-AC-1 through ARM-AC-5 from design decomposition through integration to verification closure.

Left Side of the V: Design Decomposition

RM-7 and RM-8 drove the 2-DOF coaxial differential architecture: both motors co-located at the pivot (ARM-SR-2 mass centralisation), 3:1 helical-spur gear reduction for torque multiplication (ARM-SR-1), and the dual-supply power rail separating the 24 V motor bus from the 12 V logic rail (ARM-AC-3 regenerative braking protection). ARM-AC-1 drove the motor selection journey (servo → ODrive → Damiao DM-J4310-2EC), the 200 Hz unified Teensy control loop, and the Sparse Edge-Trigger CAN strategy. ARM-AC-2 drove the thermal margin analysis (PLA Tg ~55–60°C), the PETG housing specification, and the Damiao temperature telemetry integration.

Base of the V: Integration Build

At integration stage, the coaxial differential joint was assembled (V5: stainless D-shaft, Delrin pin, M2 screw reinforcement post-CDE-Fair), calibrated via the 4-test motor probe, and integrated with the Teensy firmware and micro-ROS pipeline. The strike library was populated with all six standard strikes (Left/Right Jab, Hook, Uppercut), and the state-aware motion planner (vector alignment skip, proportional snap-back) was commissioned.

Right Side of the V: Verification Closure

ARM-AC-1 closed with 0.64 s best time at 30 rad/s (N=43 benchmark): Pass.
ARM-AC-5 closed with 0.69 A peak current (N=43): Pass.
ARM-AC-2, ARM-AC-3, ARM-AC-4 pending formal test sessions: Partial.
RM-7 closes once all six strike types are demonstrated in sustained sparring (ARM-AC-2 session). Full test procedures are in Testing & Evaluation (Section 5.2.5.5).

Acceptance Criteria Summary

The following table recaps the five measurable acceptance criteria derived from the requirements above. Torque calculations supporting ARM-AC-1 are provided in Appendix 1.

ID Criterion Target Req Link Method
ARM-AC-1 Strike speed (90° rotation) ≤ 0.70 s RM-8 Firmware timestamp delta (N=43 speed tests)
ARM-AC-2 Sparring endurance 5 min continuous, <60°C motors RM-8 Continuous sparring session + Damiao temperature telemetry
ARM-AC-3 Regenerative braking safety No PSU trip over 5 emergency stops RM-7, RM-8 Visual confirmation and voltmeter monitoring
ARM-AC-4 Multi-strike chain execution 3-strike chain ≤ 5 s total RM-7 Video timestamp and GUI console log
ARM-AC-5 Peak motor current within safety limit < 2.0 A per motor RM-8 CAN current telemetry (N=43 speed tests)

Subsection Highlights

Section Key Contribution
5.2.5.1 Design & Ideation 2-DOF coaxial differential joint selected from three motor platforms (servo, ODrive, Damiao); mass centralised at pivot to minimise moment of inertia.
5.2.5.2 Mechanical Design Structural failures observed at CDE Fair informed a material science revision from polymer shafts to a 6 mm stainless steel D-shaft with Delrin pin reinforcement, achieving sustained sparring capability.
5.2.5.3 Electrical Integration Dual-supply architecture (24 V motor bus / 12 V isolated logic rail) prevents regenerative braking OVP events from triggering a full robotic shutdown.
5.2.5.4 Firmware & Software Teensy 4.0 executes a deterministic 200 Hz unified loop (CAN + I2C + micro-ROS); a two-tier ROS 2 architecture separates real-time motor control from combat decision logic on the Jetson Orin NX.

Design Evolution

Five major iterations were required to achieve the final deployed configuration. The following table summarises each transition and the engineering rationale.

Iteration Configuration Outcome Transition Rationale
1. Interim Gear-rail coaxial joint with servo motors DOF validated; servos failed under impact load BLDC motors required for impact resilience
2. ODrive ODrive V3.6 with 360 KV outrunner FOC validated; jitter, no absolute position, excessive volume Integrated actuator required
3. Damiao DM-J4310-2EC with 3:1 external spur gearbox (30:1 total) All requirements met; CDE Fair demonstration successful Structural failures exposed at CDE Fair under repeated impact
4. Post-CDE-Fair Stainless D-shaft, Delrin pin, M2 screw reinforcement Structural integrity validated; sustained sparring achieved Stateless return-to-zero motion produced unacceptable mechanical wear and poor sparring realism
5. Final State-aware motion planning: vector alignment skip, proportional snap-back recovery Fluid, continuous sparring; reduced mechanical wear Current deployed configuration

Final Arm Design

The final deployed arm assembly consists of a compact 2-DOF coaxial differential joint, two Damiao DM-J4310-2EC integrated actuators mounted at the pivot, a 3:1 external helical-spur gear reduction stage, and a training stick with multi-layer padding incorporating embedded MPU6050 IMUs for impact detection.

Interact with the 3D model to explore the arm joint assembly. Toggle annotations to view key components. Hide Annotations
Interactive 3D Model: V5 Coaxial Differential Arm Joint Assembly. Full annotated view in Section 5.2.5.2 Mechanical Design.

Power Architecture

The arm motor power subsystem operates from a Mean Well LRS-200-24 (200 W, 24 V, 8.8 A) switched-mode PSU. All four Damiao motors receive power in parallel from a common distribution busbar. The 12 V logic rail is isolated from the motor bus to prevent regenerative braking OVP events from disrupting the Teensy or Jetson. Measured steady-state draw is approximately 8 W per motor at 24 V across 43 validated tests, with a combined peak of approximately 33 W.

Arm actuation power subsystem diagram
Arm actuation power subsystem: Mean Well LRS-200-24 (200 W, 24 V, 8.8 A) supplying four Damiao DM-J4310-2EC motors in parallel via a distribution busbar, with RegenClamp V0.3 protection against back-EMF OVP spikes. Click to enlarge.

Data Architecture

The arm actuation data path spans three hardware layers. At the microcontroller level, the Teensy 4.0 issues Position-Speed mode CAN commands to the four Damiao motors at up to 200 Hz (Sparse Edge-Trigger strategy) and polls the four MPU6050 IMUs across two I2C buses. Motor feedback and IMU data are packed into a 21-double payload and published via micro-ROS to the Jetson Orin NX. The Jetson hosts the GUI application, which executes the combat decision logic and publishes motor commands on the /motor_commands topic at 100 Hz.

Arm actuation data and control architecture
Arm actuation data and control architecture: CAN bus (1 Mbps, 4× Damiao DM-J4310-2EC), I2C dual-bus (400 kHz, 4× MPU6050 IMUs), micro-ROS USB bridge to Jetson Orin NX, and ROS 2 GUI command loop at 100 Hz. Click to enlarge.

Test and Evaluation Results

A controlled speed-test benchmark was conducted across all six standard strikes at set speeds ranging from 10 to 30 rad/s, with a minimum of three repetitions each, for a total of 43 validated tests. All tests were completed with zero safety trips, zero current-limit events, and zero position timeouts. Full methodology and raw data are documented in Section 5.2.5.5 Testing and Evaluation.

Acceptance Criteria Outcome

ID Criterion Target Measured Status
ARM-AC-1 Strike speed (90°) ≤ 0.70 s 0.64 s best (Left Jab at 30 rad/s) Pass
ARM-AC-2 Sparring endurance (5 min, <60°C) 5 min continuous Pending formal session Pending
ARM-AC-3 Regenerative braking safety No PSU trip / 5 E-stops Pending dedicated test session Pending
ARM-AC-4 Multi-strike chain (≤ 5 s for 3 strikes) ≤ 5 s total Pending chain-execution test Pending
ARM-AC-5 Peak motor current < 2.0 A per motor 0.69 A max (Right Hook, M4); N=43 Pass

Full test methodology, per-strike benchmark data, PID overhead analysis, and video evidence are documented in Section 5.2.5.5 Testing and Evaluation.

Limitations

Damiao PID Firmware Acceleration and Deceleration Ramps

The most significant current limitation of the arm actuation subsystem is not a mechanical or electrical constraint, but a firmware one inherent to the Damiao DM-J4310-2EC actuator platform. The system-level requirement RM-5 specifies that a 90° arm sweep must be completed in 0.25 s. Kinematic analysis and torque calculations confirm that the motor's rated output torque and top angular velocity are mechanically sufficient to achieve this target within the arm's moment of inertia; the physical system is not speed-limited by its mechanics. The limiting factor is the Damiao actuator's internal closed-loop PID controller, which applies a fixed acceleration ramp at the start of every commanded motion and a symmetric deceleration ramp approaching the target position, irrespective of the commanded speed setpoint. Across 43 validated speed-benchmark trials, this ramp overhead was measured at approximately 0.4 to 0.5 s per strike, constituting the dominant fraction of the total motion time.

As a consequence, the best recorded full-stroke time of 0.64 s (Left Jab at 30 rad/s) reflects the PID ramp overhead rather than a kinematic ceiling. The requirement target of 0.25 s could be approached only if the Damiao firmware were modified to reduce or eliminate its internal ramp profile, or if the actuator were replaced with a platform permitting direct torque or velocity control without internal position-loop overhead. For the purposes of this report, RM-5 is classified as partially fulfilled: the arm's kinematic design meets the intent of the requirement, and the shortfall is attributed specifically to an externally imposed firmware constraint rather than a design deficiency in the mechanical or electrical architecture. A full analysis of the PID overhead, including measured ramp profiles across all six strike types and speed setpoints, is documented in Section 5.2.5.5 (Testing and Evaluation).

Continue to Detailed Documentation

Detailed Documentation

Complete technical documentation is organised across five sections: