Project Assignment 2

Design, Build and Simulate your Bio-Inspired Robot

A foldable quadruped robot with knuckle-walking locomotion inspired by African apes

Dheeraj Reddy Pasham
Dhansukh Jaladi
Yeshwanth Voruganti

December 8, 2025

Assembled foldable quadruped robot prototype
Laser cutting process for robot components

Executive Summary

This project presents the design, optimization, and fabrication of a bio-inspired foldable quadruped robot with knuckle-walking locomotion characteristics. Inspired by African apes' adaptive gait strategies and recent advances in variable-stiffness actuation, we developed a laminate-based robot capable of forward locomotion through asymmetric leg actuation patterns.

42 Configurations

Parameter sweep testing 6×7 actuator signal combinations

0.74 m/s Max Speed

Optimal asymmetric actuation (+0.5 rad / -0.8 rad)

5-Layer Laminate

Laser-cut cardboard with plastic hinge mechanism

84% Correlation

Strong sim-to-real correlation preserved rankings

Project Videos

Project Video

Robot Motion Video

1. Introduction

1.1 Project Overview

This project investigates the design and optimization of a bio-inspired foldable quadruped robot featuring knuckle-walking locomotion patterns inspired by African apes. The robot employs laminate-based construction with origami-inspired folding mechanisms and position-controlled actuators to achieve forward locomotion.

Biological Inspiration

African apes (gorillas, chimpanzees, bonobos) have evolved knuckle-walking as a specialized terrestrial locomotion strategy that combines stable weight distribution with behavioral flexibility. This adaptive strategy allows apes to modulate contact forces, protect finger joints, and efficiently traverse diverse terrain while maintaining the capability for arboreal behaviors.

1.2 Key Design Principles

Foldable Construction

Origami-based soft robotics enables compact storage, simple fabrication, and compliant joints for passive shock absorption.

Asymmetric Actuation

Different positive and negative leg angles create forward-biased motion patterns without complex variable-stiffness joints.

Position Control

Servo-like position actuators simplify control architecture while enabling precise, repeatable leg trajectories.

1.3 Key Objectives

  1. 1Develop a dynamic MuJoCo model with material compliance and motor behavior
  2. 2Optimize asymmetric actuator signal parameters to maximize locomotion speed
  3. 3Fabricate a physical laminate prototype using laser-cut cardboard
  4. 4Document the complete design-to-fabrication workflow for educational replication

1.4 Performance Metric

Primary Metric

Maximum Horizontal Speed (m/s)

Maximum speed directly measures locomotion effectiveness and is easily comparable across different actuator configurations. The horizontal component isolates forward motion from vertical bouncing, providing a clean measure of directional propulsion efficiency.

2. Project Videos

Project Video

Robot Motion Video

2. Part 1: System Model Definition

2.1 Model Overview

The MuJoCo model represents a quadruped robot with a foldable laminate construction inspired by the original CAD design. The robot consists of a main body plate connected to four symmetric legs via compliant hinges, with a single actuator driving all legs in parallel through mechanical constraints.

Structural Components

ComponentPropertyValueNotes
Main BodyMass4.1 gCardboard laminate base plate
Dimensions63 × 50 × 5 mmRectangular base
Legs (×4)Mass each4-5 gWith compliant joints
Top PlateMass2.5 g6 DOF floating plate
Total RobotMass~12 gExcluding servo motor

2.2 Material Compliance

Joint Compliance

frictionloss=0.1

armature=0.005 rotational damping

Position Control Gain

kp=50

dampratio=1.0 for controlled stiffness

Spring Connections

solref="0.0001 1"

Soft constraints for passive compliance

2.3 CAD Design

Leg design in LibreCAD

Leg Assembly Design

LibreCAD DXF file with joint cutouts

Main body design in LibreCAD

Main Body Design

Base plate with servo mounting points

Parameter Identification

Individual team member contributions to system parameter identification for accurate MuJoCo simulation.

Dheeraj Reddy Pasham

ID: 1233405721

Servo Parameter Identification

Yeshwanth Voragunti

ID: 1238208480

Stiffness & Damping Identification

Dhansukh Jaladi

Friction Measurement

Servo Parameter Identification— Dheeraj Reddy Pasham

Methodology

The servo parameter identification combined experimental motion capture with computational parameter fitting to characterize the SG90 servo motor's dynamic behavior. The workflow included:

  1. Video capture of servo-driven arm executing a square-wave command
  2. Motion tracking using Tracker software with dual high-contrast markers
  3. Center of rotation optimization via least-squares minimization
  4. Angle extraction and unwrapping for continuous measurement
  5. MuJoCo simulation with P-controller and motor dynamics
  6. Sensitivity analysis for Kp and viscous friction parameters

Experimental Setup

An SG90 micro-servo was mounted to a rigid table with a Chapstick tube (9.08g, 67mm length) attached to the servo horn as a known inertial load. Video was captured at 30fps using a gimbal-stabilized smartphone to minimize perspective distortion.

Identified Parameters

ParameterValue
Nominal Voltage (Vnom)6 V
Gear Ratio (G)55.5
Torque Constant (kt)4.5e-3 N·m/A
Coil Resistance (R)10 Ω
Back-EMF Constant (ke)4.5e-3 V·s/rad
Viscous Friction (b)2.9e-6 N·m·s/rad
Proportional Gain (Kp)15-20
Maximum Angle Range~172°
Key Finding

The simulation trajectory shows excellent agreement with experimental data, both exhibiting the characteristic "shark-fin" shape with linear velocity ramps and rounded peaks. This validates that the P-controller model correctly captures the servo's velocity-limited behavior.

Stiffness & Damping Identification— Yeshwanth Voragunti

Goal

Determine the stiffness characteristics of the links and overall body structure of the robot constructed using cardboard components and 4-bar leg mechanisms through free vibration analysis.

Methodology

  • Cantilever beam vibration test captured with Tracker software
  • Nonlinear least-squares fit to decaying sinusoidal model
  • Young's modulus identification: E = 0.77 GPa for cardboard
  • Series/parallel combination of link stiffnesses for total system
  • Deck modeled as clamped-clamped beam

Stiffness Parameters

ComponentStiffness
Single Link12.80 N/m
One Leg (2 links series)6.40 N/m
One Side (2 legs parallel)12.80 N/m
All Legs Combined25.60 N/m
Deck (clamped-clamped)28,974.8 N/m
Whole Robot25.58 N/m

Damping Parameters

ParameterValue
Damping Ratio (ζ)0.104
Natural Frequency (ωn)19.67 rad/s
Single Link Damping (c)0.0105 N·s/m
Full Robot Damping0.0122 N·s/m
Key Finding

The deck is significantly stiffer (28,974.8 N/m) than the legs (25.60 N/m combined), making the overall robot stiffness and damping behavior primarily governed by the compliant legs rather than the rigid deck structure. Total robot stiffness: 25.58 N/m.

Friction Measurement— Dhansukh Jaladi

Goal

Measure static and dynamic friction coefficients between the foot material and floor surface using an inclined plane experiment for use in MuJoCo simulation.

Procedure

  1. Cut a sample of the foot material
  2. Place a mass of known weight on the sample
  3. Slowly increase ramp angle until block begins to slide → record θs (static)
  4. Reduce angle until block slides with constant velocity → record θk (kinetic)
  5. Compute friction coefficients: μ = tan(θ)

Identified Parameters

ParameterValue
Static Angle (θs)21.181°
Dynamic Angle (θk)18.650°
Static Friction (μs)0.3875 ± 0.0100
Kinetic Friction (μk)0.3375 ± 0.0097
Key Finding

The results demonstrate the expected hierarchy μs > μk (~13% difference), consistent with typical dry Coulomb friction. The kinetic friction value (μk = 0.3375) is used directly in MuJoCo to tune the sliding friction parameter for accurate simulation.

3. Part 2: Design Optimization

3.1 Optimization Strategy

Approach: Exhaustive Parameter Sweep (Grid Search)

Given the relatively small design space (two independent variables: positive and negative leg angles) and the need to understand performance across the entire parameter region, an exhaustive grid search is more appropriate than gradient-based methods.

Positive Signal (θ⁺)

Leg Extension

  • • Range: 0.1 - 0.6 rad (~5.7° - 34.4°)
  • • Samples: 6 values
  • • Strategy: Linear uniform spacing

Negative Signal (θ⁻)

Leg Flexion

  • • Range: -0.8 to -0.2 rad (~45.8° - 11.5°)
  • • Samples: 7 values
  • • Strategy: Linear uniform spacing

Total Configurations Tested

6 × 7 = 42

3.2 Optimization Results

Four test cases performance comparison

Best Configuration (Case 2)

0.0812 m/s

Highest average speed with 75-123° range

Best Displacement (Case 2)

0.0407 m

Maximum horizontal displacement achieved

3.3 Key Finding: Asymmetric Gaits Win

The optimal configuration achieved a maximum speed of 0.7374 m/s with asymmetric actuation (+0.5 rad / -0.8 rad), demonstrating that asymmetric "push-pull" gaits significantly outperform symmetric patterns. This finding aligns with biological observations of knuckle-walking primates that modulate limb extension asymmetrically for efficient terrestrial locomotion.

4. Part 3: Manufacturing

4.1 5-Layer Laminate Structure

The robot is constructed using a 5-layer laminate design optimized for lightweight construction with adequate structural rigidity.

LayerMaterialPurposeSpecifications
Layer 1CardboardStructural supportBase layer for rigidity
Layer 2AdhesiveBonding agentJoins cardboard to plastic
Layer 3Plastic filmHinge mechanismProvides joint compliance
Layer 4AdhesiveBonding agentJoins plastic to cardboard
Layer 5CardboardReinforcementTop layer for rigidity

4.2 Laser Cutting Process

Laser cutting cardboard

Cardboard Layer Cutting

Precision laser cutting for structural components

Laser cut adhesive components

Adhesive Layer Components

All laminate pieces ready for assembly

4.3 Actuator: SG90 Servo Motor

Basic Parameters

Nominal Voltage
6 V DC
Stall Torque
0.0027 N·m
Stall Current
600 mA
No-load Speed
~60 RPM
Gear Ratio
55.5:1
Weight
9 g

Motor Constants

Torque Constant (k_t)
0.0045 N·m/A
Speed Constant (k_v)
0.00938 V·s/rad
Back-EMF Constant
0.0045 V·s/rad
Damping (b)
1.41e-6 N·m·s/rad
Operating Voltage
3.3V (USB)
Control Method
PWM Position

4.4 Assembled Prototype

Assembled robot prototype

Final Assembled Robot

Foldable quadruped robot with 5-layer laminate construction, SG90 servo motor, and compliant plastic hinges for knuckle-walking locomotion.

5. Part 4: Experimental Validation

5.1 Experimental Setup

Primary Measurement Tool

Open Source Physics (OSP) Tracker

  • • Motion tracking and kinematic analysis
  • • Input: .mov video files at 30 fps
  • • Output: Position data and CSV exports
  • • Manual reference marker placement

Test Environment

  • • Surface: Flat, level table
  • • Dimensions: ~2m × 1m test area
  • • Temperature: Room temp (~20-25°C)
  • • Lighting: Uniform bright lighting
  • • Camera: Smartphone at 30 fps HD

5.2 Motion Tracking Process

Tracker software analyzing robot motion - Test 1

Tracker Analysis - Session 1

Position tracking with coordinate calibration

Tracker software analyzing robot motion - Test 2

Tracker Analysis - Session 2

Multiple data plots showing x, y position over time

5.3 Measurement Uncertainty

Sources of Error

Spatial Resolution

±2-3 pixels (~0.5-1 mm)

Temporal Resolution

33 ms between frames

Manual Tracking

±3-5 pixels placement error

5.4 Velocity Analysis

Robot speed analysis over time

Robot Speed Over Time

Left: Total and horizontal speed with mean velocity of 0.0388 m/s. Right: Velocity components showing X, Y, and Z velocities over the 5-second test duration.

6. Results and Discussion

6.1 Experimental Results Summary

Test CaseTotal Time (s)Displacement (m)Avg Velocity (m/s)
T1117.000.27480.2348
T2 ✓ Best129.170.32400.2508
T3196.400.28520.1452
T4132.630.25920.1954
Mean143.800.28580.2065

6.2 Hardware vs Simulation Comparison

MuJoCo Mean Velocity

0.4675 m/s

Hardware Mean Velocity

0.2066 m/s

Sim-to-Real Gap

55.8%

Critical Finding: Rankings Preserved!

Despite the 55.8% velocity gap, the ranking of configurations is perfectly preserved:

MuJoCo Ranking:

  1. T2 (0.650 m/s)
  2. T1 (0.450 m/s)
  3. T4 (0.420 m/s)
  4. T3 (0.350 m/s)

Hardware Ranking:

  1. T2 (0.251 m/s) ✓
  2. T1 (0.235 m/s) ✓
  3. T4 (0.195 m/s) ✓
  4. T3 (0.145 m/s) ✓

6.3 Sources of Sim-to-Real Gap

Unmodeled Friction

~30-40% of gap

Simulation assumed ideal joints; hardware has friction from laser-cut edges, hinge contact, leg-ground interaction.

Unmodeled Energy Losses

~10-15% of gap

Material damping in cardboard, air resistance, servo dead-band and control lag (~50ms response time).

Servo Model Simplification

~5-10% of gap

Simulation modeled servo as ideal torque source; real servo has finite torque output, saturation effects.

6.4 Key Finding: T2 Configuration is Optimal

Highest Average Velocity

0.2508 m/s

71.4% better than worst case

Maximum Displacement

32.40 cm

Optimal path achieved

Balanced Performance

129.17 s

High speed without excessive duration

7. Future Work (V2)

Design Improvements

  • • Multi-DOF leg mechanism for independent control
  • • Advanced materials (CFRP or aluminum composite)
  • • Variable stiffness joint design
  • • Expected: 30-50% speed increase

Modeling Improvements

  • • Hysteresis modeling (Kelvin-Voigt)
  • • Contact dynamics refinement
  • • Servo actuator fidelity improvements
  • • Structural elasticity modeling

Experimental Improvements

  • • High-speed motion capture (120+ fps)
  • • Force plate integration
  • • IMU-based locomotion analysis
  • • Controlled surface variation testing

Additional Studies

  • • Multi-objective optimization (3D parameter space)
  • • Terrain adaptation study
  • • Scaling laws investigation
  • • Evolutionary optimization for gait discovery

8. Conclusions

Summary of Findings

1. T2 Configuration is Optimal

Experimental validation confirmed Test T2 achieved highest average velocity (0.2508 m/s), representing 71.4% better performance than T3.

2. Significant Parameter Sensitivity

51% velocity spread across four test cases demonstrates high sensitivity to servo control parameters.

3. Robust Mechanical Design

Despite velocity variations of ±21.6%, displacement consistency remained high (±8.4% variation).

4. Simulation-Reality Gap

Strong correlation (0.8432) validates simulation for comparative optimization, despite 55.8% absolute velocity gap.

Achievement of Objectives

Objective 1 - Develop Simulation Model: Created comprehensive MuJoCo model with asymmetric servo actuation

Objective 2 - Optimize Design Parameters: Performed parameter sweep across servo angle ranges

Objective 3 - Fabricate Physical Prototype: Successfully manufactured 5-layer laminate robot

Objective 4 - Experimental Validation: Conducted four test configurations with motion tracking analysis

Objective 5 - Sim-Experiment Comparison: Partially achieved; full quantitative correlation requires additional characterization

Broader Implications

This project demonstrates that laser-cut composite structures can achieve competitive locomotion performance with minimal complexity. The 5-layer laminate approach offers:

  • • Manufacturing accessibility: Standard laser cutter, no 3D printing required
  • • Design flexibility: Easy modification of layer materials and thicknesses
  • • Educational value: Complete design-to-fabrication workflow documented
  • • Bio-inspired validation: Asymmetric gaits match biological observations