seb/saltylab-firmware
1
0
Fork 0
Code Issues Pull Requests 3 Actions Packages Projects Releases Wiki Activity
saltylab-firmware/jetson/docs/power-budget.md
sl-firmware fa75c442a7 feat: remove all STM32/Mamba/BlackPill references — ESP32-S3 only 
Archive STM32 firmware to legacy/stm32/:
- src/, include/, lib/USB_CDC/, platformio.ini, test stubs, flash_firmware.py
- test/test_battery_adc.c, test_hw_button.c, test_pid_schedule.c, test_vesc_can.c, test_can_watchdog.c
- USB_CDC_BUG.md

Rename: stm32_protocol → esp32_protocol, mamba_protocol → balance_protocol,
  stm32_cmd_node → esp32_cmd_node, stm32_cmd_params → esp32_cmd_params,
  stm32_cmd.launch.py → esp32_cmd.launch.py,
  test_stm32_protocol → test_esp32_protocol, test_stm32_cmd_node → test_esp32_cmd_node

Content cleanup across all files:
- Mamba F722S → ESP32-S3 BALANCE
- BlackPill → ESP32-S3 IO
- STM32F722/F7xx → ESP32-S3
- stm32Mode/Version/Port → esp32Mode/Version/Port
- STM32 State/Mode labels → ESP32 State/Mode
- Jetson Nano → Jetson Orin Nano Super
- /dev/stm32 → /dev/esp32
- stm32_bridge → esp32_bridge
- STM32 HAL → ESP-IDF

docs/SALTYLAB.md:
- Update "Drone FC Details" to describe ESP32-S3 BALANCE board (Waveshare ESP32-S3 Touch LCD 1.28)
- Replace verbose "Self-Balancing Control" STM32 section with brief note pointing to SAUL-TEE-SYSTEM-REFERENCE.md

TEAM.md: Update Embedded Firmware Engineer role to ESP32-S3 / ESP-IDF

No new functionality — cleanup only.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-04 09:00:38 -04:00

6.4 KiB
Raw Blame History

Jetson Orin Nano Super Power Budget Analysis

Self-Balancing Robot — 25W Envelope

Last updated: 2026-02-28 Target: Operate within 25W SoC power envelope (MAXN 25W mode)

Power Modes

Jetson Orin Nano Super supports multiple NVPModel power modes:

Mode ID CPU GPU TDP
MAXN 0 6× A78AE @ 1.5GHz 1024-core Ampere 25W
15W 1 6× A78AE @ 1.2GHz 1024-core Ampere 15W
10W 2 4× A78AE @ 1.2GHz 1024-core Ampere 10W
7W 3 4× A78AE @ 0.8GHz 1024-core Ampere 7W

For this robot, we target MAXN 25W mode — a significant upgrade from the previous Nano 10W budget.

# Check current mode
sudo nvpmodel -q

# Set 25W MAXN mode
sudo nvpmodel -m 0

# Set 15W mode (thermal / battery save)
sudo nvpmodel -m 1

# Monitor power in real time
sudo tegrastats
# or via jtop
sudo jtop

Component Power Budget

SoC (Jetson Orin Nano Super Module)

Component Idle (W) Load (W) Peak (W) Notes
CPU (6× A78AE) 1.5 6.0 8.0 ROS2, SLAM, Nav2
GPU (1024-core Ampere) 0.8 5.0 7.0 Depth processing, DNN inference
LPDDR5 RAM (8GB) 0.4 0.8 1.0
NVMe SSD (M.2) 0.2 0.5 0.8 Map storage, rosbags
Video encoder / ISP 0.0 1.5 2.5 4× IMX219 ISP processing
SoC Subtotal 2.9 13.8 19.3

Peripherals

Peripheral Idle (W) Active (W) Peak (W) Interface Notes
RealSense D435i 0.3 1.5 3.5 USB 3.1 Peak during boot/init
RPLIDAR A1M8 0.4 2.6 3.0 USB (UART adapter) Motor spinning
<<<<<<< HEAD
ESP32 bridge 0.0 0.0 0.0 USB CDC Self-powered from robot 5V

======= | ESP32-S3 bridge | 0.0 | 0.0 | 0.0 | USB Serial (CH343) | Self-powered from robot 5V |

291dd68 (feat: remove all STM32/Mamba/BlackPill references — ESP32-S3 only) | 4× IMX219 cameras | 0.2 | 2.0 | 2.4 | MIPI CSI-2 | ~0.5W per camera active | | Peripheral Subtotal | 0.9 | 6.1 | 8.9 | | |

Total System (from Jetson 5V barrel jack)

Scenario SoC (W) Peripherals (W) Total (W) Margin vs 25W
Idle 2.9 0.9 3.8 +21.2W
Nominal (SLAM + cameras) 13.8 6.1 19.9 +5.1W ✅
Peak (DNN + all sensors) 19.3 8.9 28.2 -3.2W ⚠️

Budget Analysis vs Previous Platform

Metric Jetson Orin Nano Super Jetson Orin Nano Super
TDP 10W 25W
CPU 4× Cortex-A57 @ 1.43GHz 6× A78AE @ 1.5GHz
GPU 128-core Maxwell 1024-core Ampere
RAM 4GB LPDDR4 8GB LPDDR5
AI TOPS ~0.5 67
Nominal load 11.4W (over budget) 19.9W (5W headroom)
Cameras 0 CSI 4× IMX219 CSI
Storage microSD NVMe M.2

The Orin Nano Super has 2.5× more thermal headroom at nominal load. No aggressive power-gating needed for normal operation.

Power Compliance Strategy

Nominal Operation (SLAM + cameras) — ✅ Within 25W

At 19.9W nominal, we have 5W headroom. No mitigation required for normal robot operation.

Peak Operation (DNN inference) — ⚠️ Briefly exceeds 25W

When running DNN inference (e.g., object detection) simultaneously with full sensor suite:

Mitigation 1: Thermal throttling (automatic) The Orin's DVFS will automatically throttle CPU/GPU when temperature exceeds threshold. No explicit action needed — the Orin handles this gracefully.

Mitigation 2: Switch to 15W mode during high-load phases

sudo nvpmodel -m 1   # 15W mode: reduces peak to ~22W
sudo nvpmodel -m 0   # return to MAXN when cooling

Mitigation 3: RPLIDAR motor gating Stop RPLIDAR motor between scan cycles: saves ~2.2W average. Handled automatically by rplidar_ros driver via DTR line control.

Mitigation 4: Camera resolution reduction For compute-heavy phases, drop from 640×480 to 424×240 per camera: saves ~0.6W.

CSI Camera Bandwidth

4× IMX219 cameras at 640×480@30fps:

Parameter Value
Per-camera raw bandwidth 640×480×30×10bpp = 92.16 Mb/s
Total 4 cameras ~369 Mb/s
MIPI CSI-2 capacity (Orin) 40 Gb/s total (2× 4-lane)
ISP processing overhead ~1.5W (all 4 cameras active)

CSI bandwidth is well within capacity. The Orin Nano Super's ISP handles 4 cameras simultaneously.

Input Power Requirements

Jetson Orin Nano Super Power Input

Spec Value
Input connector 5.5mm / 2.5mm barrel jack
Input voltage 5V DC
Recommended current ≥6A (30W supply for headroom)
Absolute max 5.25V

Use a 5V 6A supply minimum. A 4A supply may brownout under DNN peak load.

Robot Power Architecture (Recommended)

LiPo 4S (16.8V max)
       │
       ├─► DC-DC Buck → 5V 6A ──► Jetson Orin barrel jack (30W)
       │   (e.g., XL4016E1)
       │
<<<<<<< HEAD
       ├─► DC-DC Buck → 5V 3A ──► ESP32 + logic 5V rail
=======
       ├─► DC-DC Buck → 5V 3A ──► ESP32-S3 + logic 5V rail
>>>>>>> 291dd68 (feat: remove all STM32/Mamba/BlackPill references — ESP32-S3 only)
       │
       └─► Hoverboard ESC ──► Hub motors (48V loop)

Using a 4S LiPo (vs 3S previously) gives better efficiency for the 5V buck converter at Orin's higher power draw.

Real-Time Monitoring

# Live power telemetry (tegrastats)
sudo tegrastats --interval 500

# Key fields:
# POM_5V_IN X/Y   — total input power (current mW / average mW)
# POM_5V_GPU X/Y  — GPU power
# POM_5V_CPU X/Y  — CPU power

# Interactive monitoring (jtop — recommended)
sudo pip3 install jetson-stats
sudo jtop

# Log power to file
sudo tegrastats --interval 1000 --logfile /tmp/power_log.txt &

# Parse log
grep "POM_5V_IN" /tmp/power_log.txt | \
  awk '{for(i=1;i<=NF;i++) if($i=="POM_5V_IN") print $(i+1)}' | \
  awk -F'/' '{sum+=$1; count++} END {print "Avg:", sum/count/1000, "W"}'

Summary

Metric Value
Target envelope 25W (MAXN)
Nominal (SLAM + 4 cameras) ~19.9W
Peak (DNN inference) ~28.2W (briefly)
Compliant scenario All sensors + SLAM (no DNN)
Recommended PSU 5V 6A (30W)
Power mode nvpmodel MAXN (Mode 0)
Upgrade from Nano +150% TDP, +13,300% AI TOPS