Project RoboSnomo

Layer 1: The Brain - NVIDIA Jetson Orin Nano Super

Compute Specifications

Camera & Vision Interfaces

MIPI CSI-2 Connectors: 2x 22-pin connectors supporting up to 6 simultaneous camera streams

• CAM0: 1x 2-lane CSI interface
• CAM1: 1x 4-lane CSI interface (configurable as 2x 2-lane)
• Maximum Resolution: 4K @ 90 FPS with compatible sensors
• Image Signal Processor: ISP 6.x with hardware-accelerated debayering, HDR, noise reduction

I/O Capabilities

• USB: 1x USB-C 3.2 (10 Gbps) + 4x USB-A 3.2 ports
• Ethernet: 1x Gigabit Ethernet (10/100/1000 Mbps)
• GPIO: 40-pin expansion header (Raspberry Pi compatible layout) with I2C, SPI, UART, PWM
• PCIe: Gen 3 support (1x x4 + 1x x2 or 2x x1 lanes)
• DisplayPort: 1x DP 1.2 output (4K @ 30 Hz max)

CAN Bus Integration

Implementation: MCP2515 SPI-to-CAN controller + MCP2551 transceiver

Software Stack: SocketCAN Linux driver, python-can library for application layer

Environmental Specifications

• Operating Temperature: -25°C to +90°C (junction temperature)
• Recommended Range: 0°C to 50°C for standard applications
• Winter Operation: Suitable for snowmobile use with proper enclosure and thermal management
• Cooling: Active cooling (fan) required for sustained 25W MaxN mode

Software & AI Framework Support

• JetPack SDK: Version 6.2 (Ubuntu 22.04 L4T base)
• CUDA: Version 12.6 with cuDNN and cuBLAS libraries
• Deep Learning: TensorFlow 2.15, PyTorch 2.x, TensorRT optimization
• Computer Vision: OpenCV with CUDA acceleration, VisionWorks SDK, DeepStream
• ROS Support: ROS 2 Humble for robotics middleware (sensor fusion, path planning)

Layer 2: The Heart - Speeduino ECM

Hardware Platform

Recommended: STM32F407-based Speeduino (not Arduino Mega) for native dual CAN bus support

2-Stroke Engine Configuration

Challenge: Speeduino is optimized for 4-stroke engines. 2-stroke single-cylinder requires configuration workarounds.

Recommended Configuration

• Displacement: Enter half of actual engine displacement (e.g., 100cc for 200cc engine)
• Cylinders: Set to 2 (even though it's single-cylinder)
• Stroke: 2-stroke mode
• Injectors: 1 (single injector)
• Spark Mode: Single channel or wasted spark
• Load Algorithm: Alpha-N (TPS-based) recommended over Speed Density due to expansion chamber pressure pulses

Rationale: Most 2-stroke single-cylinder engines provide 1 trigger pulse per crankshaft revolution (360°). Speeduino's "Basic Distributor" mode expects 1 pulse per 720°, causing RPM to read at 2x actual speed. Configuring as a 2-cylinder engine compensates for this.

Sensor Inputs

Sensor	Type	Voltage Range	Purpose
CLT (Coolant Temp)	NTC Thermistor	0-5V analog	Engine protection, warmup enrichment
IAT (Intake Air Temp)	NTC Thermistor	0-5V analog	Density compensation, fuel corrections
TPS (Throttle Position)	3-wire Potentiometer	0-5V analog	Alpha-N load source, acceleration detection
RPM/Crank Position	Hall Effect or VR	0-5V or 0-12V digital	Engine speed, ignition timing reference
O2 Sensor (Optional)	Wideband Controller	0-5V analog	Closed-loop fuel tuning, autotune

Fuel & Ignition Management

Fuel Mapping

• VE Table Resolution: 16x16 grid (256 cells)
• Axes: RPM (X-axis) vs Load (Y-axis, TPS % for Alpha-N)
• Interpolation: 3D interpolation for smooth transitions between cells
• Corrections: Coolant temp, air temp, acceleration enrichment, cranking enrichment

Ignition Timing

• Advance Table: 16x16 grid matching fuel table axes
• Range: Typically -10° to +40° BTDC (user-configurable)
• Dwell Control: Variable dwell map with battery voltage compensation
• CDI Integration: Use 5V logic output to trigger stock CDI (maintains proven high-energy spark)

CAN Bus Communication

Protocol: OBD2 standard PIDs + custom message definitions

Telemetry to Jetson (100 Hz):

• Engine RPM (PID 0x0C)
• Coolant temperature (PID 0x05)
• Intake air temperature (PID 0x0F)
• Throttle position (PID 0x11)
• Battery voltage, error codes, injector duty cycle

Commands from Jetson:

• Throttle position setpoint (autonomous mode)
• Engine enable/disable (safety override)

TunerStudio Configuration

All ECM parameters are configured via TunerStudio MS software (Windows/Mac/Linux):

• Connection: Serial over USB (115200 baud)
• Real-Time Tuning: Modify fuel/ignition tables while engine running
• Autotune: Laptop-based algorithm automatically adjusts VE table using wideband O2 feedback
• Data Logging: 10-20 samples/second for post-analysis

Layer 3: The Muscle - Actuation Systems

Steering Actuation

Recommended: Progressive Automations PA-04-12-400

Professional Upgrade Option: LINAK LA77 or TiMOTION MA2 (6000-10000N force, IP69K rated, CAN bus interface, -40°C to +85°C, $600-1500)

Brake Actuation

Recommended: Progressive Automations PA-04-6-400

Force Calculation: Target brake pressure of 1200 PSI requires ~530 lbf (2360N). A 400 lb actuator is adequate for most scenarios; 800-1000 lb provides safety margin for emergency stops.

Safety Feature: Consider spring-applied hydraulic release (SAHR) brake system for true fail-safe operation—automatic braking on power loss.

Throttle Control

High-Performance: Dynamixel MX-106T

Budget Alternative: Hitec HS-7950TH RC servo (32 kg·cm torque, standard RC PWM, $100-120, weatherproof with enclosure)

Note: Dynamixel rated -5°C to +80°C; requires heated enclosure for extreme cold snowmobile operation.

Control Interfaces

Interface	Best For	Advantages	Disadvantages
CAN Bus / CANopen	Professional multi-actuator system	Position control, diagnostics, standardized protocol	Requires CAN-enabled actuators (higher cost)
RS-485	Dynamixel servo motors	Long-distance, robust, multiple devices on one bus	Requires Dynamixel protocol library
PWM	Budget RC servos, proof-of-concept	Simple, widely supported	No built-in feedback (requires separate sensors)
Analog (0-5V or 0-10V)	Industrial actuators without digital interface	Simple wiring, reliable	No protocol-based diagnostics

Power Requirements

Peak Current Draw (All Systems):

• Steering actuator: 10-12A peak
• Brake actuator: 10-12A peak
• Throttle servo: 2-5A peak
• Jetson Orin Nano: 5A typical (25W MaxN mode @ 12V with DC-DC converter)
• Total System: ~37A peak draw

Battery Recommendation:

• Budget: 60Ah AGM battery (Optima YellowTop D31A, $250-300)
• Performance: 50Ah LiFePO4 with heating blanket ($600-800) for better cold-weather performance
• Cold Weather Adjustment: Battery capacity drops 20-40% at -20°C; size accordingly

Mounting & Mechanical Integration

• Linkage Type: Clevis (dual-pivot) mounting most common for linear actuators
• Rod Ends: Use quality spherical bearings (robosnomo COM-series, QA1 Precision) to minimize backlash
• Backlash Target: <5mm at ski for steering precision
• Mechanical Advantage: Optimize linkage geometry to reduce actuator force requirements
• Vibration Damping: Rubber bushings at pivot points, isolate electronics with Lord Micromounts
• Materials: Aluminum 6061-T6 (lightweight, corrosion-resistant), stainless steel for high-stress applications

CAN Bus Integration & Communication

Protocol Specifications

Physical Layer

Differential Signaling

• Recessive State (Logic 1): CAN-H = 2.5V, CAN-L = 2.5V, Differential = 0V
• Dominant State (Logic 0): CAN-H = 3.5V, CAN-L = 1.5V, Differential = 2.0V
• Principle: Dominant (0) overrides recessive (1), enabling priority-based arbitration

Wiring Requirements

• Cable Type: Shielded twisted pair, 120Ω characteristic impedance
• Twist Rate: Minimum 40 twists/meter (ideally 1 twist per 25mm)
• Wire Gauge: 20-24 AWG for automotive applications
• Recommended Cable: Belden 3084A, DeviceNet cable, or automotive-grade CAN cable
• Termination: 120Ω resistors at both ends of bus only (measure 60Ω total)
• Stub Length: Keep actuator/sensor connections under 30cm maximum
• Shielding: Ground shield at ONE point only to prevent ground loops

Network Topology

Message ID Architecture (Priority-Based)

ID Range	Subsystem	Priority	Examples
0x001-0x0FF	Safety & Emergency	Highest	Emergency stop (0x001), Kill switch (0x002)
0x100-0x1FF	ECM Critical Data	High	Engine RPM (0x100), Coolant temp (0x110)
0x200-0x2FF	Actuator Commands	High	Steering (0x200), Brake (0x210), Throttle (0x220)
0x300-0x3FF	Vision/Sensor Data	Medium	Obstacle detection (0x300), GPS (0x310)
0x400-0x4FF	Telemetry/Diagnostics	Low	Status messages (0x400), Logs (0x410)
0x500-0x7FF	Future Expansion	Variable	System heartbeat (0x500), Reserved

Note: Lower ID numbers have higher priority. During arbitration, ID 0x001 will always win over ID 0x100.

Message Timing & Frequency

Message Type	ID	Frequency	Sender	Receivers
Emergency Stop	0x001	Event-driven	Any node	All
Engine RPM	0x100	100 Hz (10ms)	Speeduino	Jetson, Android App
Steering Command	0x200	50 Hz (20ms)	Jetson	Steering Actuator
Brake Command	0x210	50 Hz (20ms)	Jetson	Brake Actuator
Obstacle Detected	0x300	Event-driven	Jetson	All
System Heartbeat	0x500	10 Hz (100ms)	Jetson	All actuators

Error Handling & Fault Confinement

CAN bus includes five built-in error detection mechanisms:

1. Bit Error: Node detects mismatch between transmitted and received bit
2. Stuff Error: More than 5 consecutive identical bits (violates bit stuffing rule)
3. CRC Error: Cyclic redundancy check fails
4. Form Error: Fixed-format fields contain illegal values
5. Acknowledgment Error: No node acknowledges message reception

Three Error States

• Error Active: Normal operation (TEC < 128 && REC < 128), sends active error frames
• Error Passive: Degraded mode (128 ≤ TEC or REC < 256), sends passive error frames
• Bus-Off: Node disconnected (TEC ≥ 256), requires recovery sequence

Safety Feature: Faulty nodes automatically remove themselves from the bus before disrupting communication.

Safety Heartbeat Implementation

Concept: Jetson sends periodic "alive" message every 100ms. Actuators expect heartbeat; if missing >300ms, enter safe mode (stop movement, engage brake).

Actuator Response: If heartbeat timeout exceeds 300ms, actuator controller immediately stops all motors, engages brake, and enters fail-safe mode until heartbeat resumes.

Vision System & AI Models

Stereo Camera System

Recommended: Arducam IMX477 Stereo Camera Kit

Vision Processing Pipeline

1. Image Acquisition: Synchronized stereo frame capture at 60 FPS (720p for real-time performance)
2. Rectification: Transform images to align epipolar lines (one-time calibration matrix applied)
3. Stereo Matching: SGBM (Semi-Global Block Matching) algorithm generates disparity map
4. Depth Estimation: Convert disparity to 3D point cloud (depth = baseline × focal_length / disparity)
5. Object Detection: YOLOv8 identifies obstacles (trees, rocks, other vehicles)
6. Semantic Segmentation: Classify terrain (snow, ice, trail, vegetation)

Performance: End-to-end latency of 44-78ms for full 1080p pipeline. Using 720p resolution achieves 60+ FPS for real-time autonomous control.

Object Detection Models

Recommended: YOLOv8n (Nano) with INT8 Quantization

Custom Training: Retrain on winter-specific dataset (15,000+ images of snow, trees, rocks, trails, other snowmobiles). Use RSOD (Remote Sensing Object Detection) dataset as starting point.

Semantic Segmentation

Recommended: MFA-DeepLabv3+ (Multi-Feature Aggregation)

• FPS on Jetson Orin Nano: 36.4 FPS (real-time capable)
• Purpose: Classify each pixel as snow, ice, rock, vegetation, trail, sky
• Integration: Combines with stereo depth for 3D terrain understanding
• Training: Use Cityscapes dataset augmented with winter off-road images

Path Planning Algorithms

Hybrid Approach (Recommended)

• Global Planner: A* algorithm on GPS waypoint map (runs at 1 Hz, generates optimal route)
• Local Planner: DWA (Dynamic Window Approach) for real-time obstacle avoidance (10-20 Hz)
• Rationale: A* provides optimal long-range path; DWA handles dynamic obstacles detected by vision

Alternative: RRT (Rapidly-exploring Random Tree)

• Speed: 0.23s planning time (very fast)
• Trade-off: Sub-optimal paths (longer routes)
• Use Case: Fallback when A* takes too long in complex environments

Sensor Fusion Architecture

Fusion Rate: 100 Hz using Extended Kalman Filter (EKF)

Sensors Integrated:

• Stereo Vision: 3D obstacle positions, terrain classification (20 Hz)
• 77GHz Radar: Long-range obstacle detection (200m range), penetrates snow/fog (10 Hz)
• IMU (BNO085): Acceleration, gyroscope, magnetometer (100 Hz)
• GPS (ZED-F9P RTK): High-precision positioning (10 Hz, <2cm accuracy with RTK correction)

Software: ROS 2 robot_localization package implements EKF fusion, outputs fused pose estimate at 100 Hz.

Radar Sensor (Recommended)

Budget Option: Texas Instruments AWR1843 (77 GHz)

• Frequency: 77 GHz (20x better range resolution than 24 GHz)
• Range: 200-300 meters
• Range Resolution: 4 cm (vs 75 cm for 24 GHz)
• Snow/Fog Performance: Radar >> LiDAR > Camera in adverse weather
• Interface: UART/SPI to Jetson
• Cost: ~$300 (development kit)

Edge AI Optimization

TensorRT Optimization Pipeline

1. Train model in PyTorch/TensorFlow
2. Export to ONNX format
3. Convert to TensorRT engine with INT8 calibration
4. Benchmark: Measure FPS and accuracy on Jetson
5. Deploy: Load TensorRT engine in Python/C++ inference code

Performance Gains: 3-15x speedup depending on model and precision (FP32 → FP16 → INT8)

INT8 Quantization Benefits

• Speed: 2-4x faster inference vs FP32
• Power: 16% reduction in energy consumption
• Accuracy: Typically <3% mAP loss for well-calibrated models
• Memory: 4x smaller model size (fits in cache, reduces bandwidth)

Winter-Specific Challenges & Solutions

Challenge	Impact	Solution
Low Contrast (White Snow)	Stereo matching struggles with textureless surfaces	Use HDR imaging (IMX477 supports 60 FPS HDR), radar fusion
Sun Reflections on Snow	Overexposure, false obstacle detection	HDR capture, polarizing filters, adaptive exposure control
Hidden Obstacles (Under Snow)	Vision cannot detect buried rocks/logs	Radar penetrates snow surface, combines with GPS trail map
Variable Lighting (Sunrise/Sunset)	Inconsistent detection performance	Train models on diverse lighting conditions, use normalized image preprocessing
Falling Snow/Fog	Obscures vision, false positives	Radar primary sensor in poor visibility, reduce speed in degraded mode

Complete Bill of Materials

Layer 1: AI & Perception

Component	Model / Specification	Quantity	Est. Cost	Status
Main Computer	NVIDIA Jetson Orin Nano Super (1TB)	1	$599	Acquired
Stereo Cameras	Arducam IMX477 Stereo Kit (12.3MP, HDR)	1 kit	$250	Sourcing
Radar Sensor	Texas Instruments AWR1843 (77 GHz)	1	$300	Planned
IMU	BNO085 9-DOF (100 Hz fusion)	1	$20	Planned
GPS (RTK)	u-blox ZED-F9P (<2cm accuracy)	1	$200	Planned
CAN Controller	MCP2515 SPI-to-CAN + MCP2551 Transceiver	1-2	$15 ea	Planned
Enclosure	Weatherproof IP65+ (Pelican or similar)	1	$60	Planned

Layer 2: Engine Control

Component	Model / Specification	Quantity	Est. Cost	Status
ECM Board	Speeduino v0.4 (STM32F407-based preferred)	1	$150	Sourcing
CAN Transceiver	MCP2551 or TJA1050 (if not on board)	1	$5	Planned
Coolant Temp Sensor	NTC Thermistor (automotive-grade)	1	$15	Planned
Intake Air Temp Sensor	NTC Thermistor	1	$15	Planned
Throttle Position Sensor	3-wire potentiometer (0-5V)	1	$20	Planned
Wideband O2 Sensor	AEM 30-0300 X-Series + Controller	1	$200	Planned
Fuel Injector	High-Z (12-14Ω), sized for engine displacement	1	$50	Planned

Layer 3: Actuation

Component	Model / Specification	Quantity	Est. Cost	Status
Steering Actuator	Progressive Automations PA-04-12-400 (400 lbs, 12", IP66)	1	$450	Sourcing
Brake Actuator	Progressive Automations PA-04-6-400 (400 lbs, 6")	1	$380	Sourcing
Throttle Servo	Dynamixel MX-106T (84 kg·cm, RS-485)	1	$480	Planned
Position Sensors	Redundant Hall effect + Potentiometer per actuator	3 sets	$60	Planned
Rod Ends / Clevis	robosnomo COM-series spherical bearings	6	$120	Planned

Communication & Wiring

Component	Specification	Quantity	Est. Cost
CAN Bus Cable	Belden 3084A or DeviceNet, 120Ω shielded twisted pair	10 meters	$50
120Ω Termination Resistors	1/4W, 1% tolerance	4 (2 spares)	$2
Weatherproof Connectors	Deutsch DT or TE Superseal series	10	$50
Wiring Harness Materials	18-20 AWG wire, heat shrink, split loom	1 set	$100

Power System

Component	Specification	Quantity	Est. Cost
Battery	60Ah AGM (Optima YellowTop D31A) or 50Ah LiFePO4	1	$280 / $700
DC-DC Converter (Jetson)	12V to 19V, 5A+ rated (for Jetson power)	1	$40
Fuses & Circuit Breakers	Automotive blade fuses, 15A-30A	6	$30
Kill Switch (Hardware)	Emergency stop relay, latching	1	$40

Cost Summary

Note: Costs exclude vintage snowmobile chassis, mechanical fabrication materials, and tools. Budget assumes hobbyist-plus components; professional-grade system with LINAK actuators and LiFePO4 battery would be ~$5,500-6,000.

Safety Systems & Testing Protocols

Critical Safety Features

1. Hardware Emergency Stop (Kill Switch)

• Implementation: Latching relay wired directly to ignition circuit, independent of all software
• Activation: Physical button accessible to operator, wireless remote backup
• Response Time: Immediate ignition cut (<10ms electrical propagation)
• Failsafe: Power loss opens relay, killing engine (fail-safe state)
• Manual Override: Retained manual brake lever and throttle cable for human takeover

2. CAN Bus Watchdog & Heartbeat

• Jetson Heartbeat: Broadcast 0x500 message every 100ms (kernel-managed for reliability)
• Actuator Timeout: If no heartbeat received for >300ms, enter fail-safe mode
• Fail-Safe Actions: Stop all actuator movement, engage brake, center steering, throttle to idle
• Recovery: Once heartbeat resumes for >1 second, allow autonomous control to resume

3. Redundant Position Sensors

• Dual Sensors per Actuator: Hall effect encoder + analog potentiometer
• Cross-Check: If sensors disagree by >5%, flag fault and enter limp mode
• Rationale: Prevents runaway actuator due to single sensor failure

4. Vision System Failure Detection

• Frame Loss Monitor: If camera FPS drops below 20 for >2 seconds, declare vision failure
• Radar Fallback: Switch to radar-only navigation in degraded visual mode (reduced speed)
• GPS Trail Mode: Follow pre-mapped GPS waypoints if both vision and radar compromised
• Complete Failure: If all sensors fail, execute emergency stop sequence

5. Thermal & Power Monitoring

• Jetson Temperature: Monitor GPU/CPU temps; throttle AI workload if >80°C, shut down if >95°C
• Actuator Overcurrent: Monitor motor current; stop if >120% rated current for >5 seconds
• Battery Voltage: Alert at <11.5V, enter low-power mode at <11.0V, emergency stop at <10.5V

6. Geofencing (Software Boundaries)

• GPS Perimeter: Define allowed operating area as polygon on map
• Boundary Action: If GPS position exits perimeter, stop forward movement and alert operator
• Use Case: Prevent autonomous operation on roads, private property, or hazardous areas

Testing Protocols

Phase 1: Bench Testing (All Systems)

1. Power-On Test: Verify all voltage rails, check for shorts, measure current draw
2. CAN Bus Communication: Use candump/cansend to verify message transmission/reception
3. Sensor Validation: Read all sensor values, verify ranges (CLT, IAT, TPS, position feedback)
4. Actuator Stroke Test: Command full extension/retraction, verify position feedback accuracy
5. Emergency Stop Test: Trigger kill switch, verify immediate ignition cut and actuator stop
6. Jetson AI Performance: Run YOLOv8 inference, measure FPS and latency with test images

Success Criteria: All sensors reading within expected ranges, actuators respond to commands within 50ms, zero CAN bus errors over 1 hour operation, emergency stop <100ms response.

Phase 2: Static Engine Testing (Non-Moving)

1. Speeduino Startup: Power on ECM, verify TunerStudio connection, check sensor readings
2. Fuel System Priming: Test injector firing (engine not running), verify spray pattern
3. Ignition Timing: Crank engine, use timing light to verify timing marks align with TunerStudio setting
4. First Start: Use conservative fuel/timing maps, start engine, verify idle stability
5. Sensor Validation: Rev engine, confirm RPM, CLT, TPS readings update correctly on CAN bus
6. Autonomous Throttle Test: Command throttle position via CAN, verify servo response (engine running, stationary)

Success Criteria: Engine starts reliably, idles smoothly, all telemetry accurate, autonomous throttle control responsive without oscillation.

Phase 3: Tethered Testing (Limited Range)

1. Steering Actuation: Test steering commands at standstill, verify ski movement matches command
2. Brake Actuation: Test brake engagement, measure stopping force with scale
3. Low-Speed Movement: Tether snowmobile to fixed point (10m rope), test forward/reverse/steering
4. Vision Pipeline: Run stereo cameras + YOLO, verify object detection of obstacles in field of view
5. Heartbeat Timeout Test: Disconnect Jetson CAN connection mid-operation, verify actuators enter fail-safe

Success Criteria: Actuators respond correctly to commands, vision system detects obstacles, fail-safe engages within 300ms of heartbeat loss.

Phase 4: Low-Speed Autonomous Testing

1. Open Area Test: Flat snow field, no obstacles, clear weather
2. Speed Limit: Maximum 5 mph (8 km/h) for initial tests
3. Observer Present: Operator with kill switch remote walks alongside
4. Waypoint Navigation: Follow simple GPS path (straight line, gentle turns)
5. Obstacle Avoidance: Place cones in path, verify autonomous swerve or stop behavior
6. Data Logging: Record all CAN messages, vision detections, GPS positions for post-analysis

Success Criteria: Completes waypoint path without human intervention, avoids all obstacles, stays within 2m of intended GPS path, zero safety system false triggers.

Phase 5: Progressive Speed Increases

1. Incremental Speed: 5 mph → 10 mph → 15 mph → 20 mph over multiple test sessions
2. Varied Terrain: Test on flat, gentle slopes, rough trails (progressively more challenging)
3. Weather Conditions: Test in clear conditions first, then light snow, then poor visibility
4. Emergency Stop Testing: Command emergency stop at each speed increment, measure stopping distance
5. Sensor Fusion Validation: Verify radar and vision agree on obstacle positions

Success Criteria: Maintain stable control at each speed tier, obstacle detection reliable at increased speeds, stopping distance within safe limits for terrain.

Phase 6: Edge Case & Failure Mode Testing

1. Single Sensor Failures: Disconnect camera, GPS, radar individually; verify graceful degradation
2. Low Battery: Run until battery voltage triggers low-power mode, verify safe shutdown
3. Extreme Cold: Test at -20°C to verify all components operational (actuators, battery, Jetson)
4. Communication Loss: Simulate CAN bus wire disconnect, verify fail-safe engagement
5. Mechanical Binding: Simulate actuator obstruction (steering blocked), verify overcurrent detection

Success Criteria: System enters safe mode for all single-point failures, no runaway behavior, operator alerted to fault via telemetry.

Safety Margins & Conservative Design

• Actuator Force: Size actuators for 2x calculated force requirement (safety margin for icy conditions)
• Speed Limits: Initial autonomous operation limited to 50% of maximum snowmobile speed
• Obstacle Detection Range: Maintain 3x stopping distance margin (e.g., at 20 mph with 10m stop, detect obstacles at 30m)
• Battery Capacity: Never discharge below 50% SoC to preserve battery life and maintain reserve
• CAN Bus Utilization: Target <50% bus load (far below 80% limit) for timing headroom

Regulatory & Insurance Considerations

• Off-Road Only: Not legal for on-road operation in most jurisdictions (no DOT approval)
• Private Property: Conduct all testing on private land with owner permission
• Liability Insurance: Consult with insurance provider; standard policies may not cover autonomous modifications
• Safety Signage: Clearly mark vehicle as "AUTONOMOUS - TESTING IN PROGRESS"
• Documentation: Maintain detailed logs of all tests, failures, and design changes for liability protection