＜span class=“js_title_inner“＞手把手在1GHz单片机上支持大模型对话 | 技术集结＜/span＞-深圳市維司達科技有限公司

本文将演示如何基于RT-Thread，在RA8P1 Titan Board上，从设备驱动、W-iFi联网到 AIOT大模型部署，构建一个端侧AIoT应用。内容聚焦具体操作与代码实现，为在高性能MCU上进行全栈开发的工程师提供一份可直接复现的实践参考。

环境搭建

烧录方法

RT-Thread 设备驱动

Wi-Fi 联网实验

AIOT大模型对话实验

效果演示

获取同款硬件

1 环境搭建

XShell 软件：https://cdn.netsarang.net/v8/Xshell-latest-p

RT-Thread Studio 安装包：https://download-redirect.rt-thread.org/download/studio/RT-Thread_Studio_2.2.9-setup-x86_64_202412161335.exe

Titan Board SDK环境安装：https://www.rt-thread.org/document/site/#/rt-thread-version/rt-thread-standard/hw-board/ra8p1-titan-board/ra8p1-titan-board?id=rt-thread-studio-%e5%bc%80%e5%8f%91%e6%ad%a5%e9%aa%a4

FSP-6.0安装包：https://github.com/renesas/fsp/releases/download/v6.0.0/setup_fsp_v6_0_0_rasc_v2025-04.1.exe

如果网络不好，有离线SDK包（sdk-bsp-ra8p1-titan-board-1.0.0.zip），需要如下方式进行导入：

2 烧录方法

设备名称：R7KA8P1KF_CPU0

3 RT-Thread 设备驱动

实现一个RT-Thread设备

staticrt_err_tdrv_test_init(rt_device_t dev){ rt_kprintf("test drv init\n"); return RT_EOK;} staticrt_err_tdrv_test_open(rt_device_t dev, rt_uint16_t oflag){ rt_kprintf("test drv open flag = %d\n", oflag); return RT_EOK;} staticrt_err_tdrv_test_close(rt_device_t dev){ rt_kprintf("test drv close\n"); return RT_EOK;} staticrt_ssize_tdrv_test_read(rt_device_t dev, rt_off_t pos, void *buffer, rt_size_t size){ rt_kprintf("test drv read pos = %d, size = %d\n", pos, size); return size;} staticrt_ssize_tdrv_test_write(rt_device_t dev, rt_off_t pos, constvoid *buffer, rt_size_t size){ rt_kprintf("test drv write pos = %d, size = %d\n", pos, size); return size;} staticrt_err_tdrv_test_control(rt_device_t dev, int cmd, void *args){ rt_kprintf("test drv control cmd = %d\n", cmd); return RT_EOK;} intrt_drv_test_init(void){ rt_device_t test_dev = rt_device_create(RT_Device_Class_Char, 0); if(!test_dev) { rt_kprintf("test drv create failed!"); return -RT_ERROR; } test_dev->init = drv_test_init; test_dev->open = drv_test_open; test_dev->close = drv_test_close; test_dev->read = drv_test_read; test_dev->write = drv_test_write; test_dev->control = drv_test_control; if(rt_device_register(test_dev, "test_drv", RT_DEVICE_FLAG_RDWR) != RT_EOK) { rt_kprintf("test drv register failed!"); return -RT_ERROR; } return RT_EOK;}MSH_CMD_EXPORT(rt_drv_test_init, enable test drv app); staticintdrv_test_app(void){ rt_device_t test_dev = rt_device_find("test_drv"); if(test_dev == RT_NULL) { rt_kprintf("can not find test drv!"); return -RT_ERROR; } rt_device_open(test_dev, RT_DEVICE_OFLAG_RDWR); rt_device_control(test_dev, RT_DEVICE_CTRL_CONFIG, RT_NULL); rt_device_write(test_dev, 100, RT_NULL, 1024); rt_device_read(test_dev, 20, RT_NULL, 128); rt_device_close(test_dev); return RT_EOK;}MSH_CMD_EXPORT(drv_test_app, enable test drv app);

PIN 设备解析

4Wi-Fi 联网实验

1.首先我们需要传输wifi的固件资源到文件系统：

参考：https://rt-thread-studio.github.io/sdk-bsp-ra8p1-titan-board/latest/driver/Titan_driver_wifi/README_zh.html#id11

2.在MSH命令行中，输入以下命令扫描附近的wifi热点信息：

wifi scan

3.在MSH命令行中，输入以下命令来让wifi模块联网：

wifi join [wifi 名称] [wifi 密码]

4.我们可以使用ping命令测试是否联网成功：

5.添加自动联网应用代码

#include<rtthread.h>#include<board.h>#include"hal_data.h"#include<fal.h>#include<dfs_fs.h>#include<wlan_mgnt.h>#include<wlan_prot.h>#include<wlan_cfg.h>#define LOG_TAG "wifi"#include<drv_log.h>#define LED_PIN_0 BSP_IO_PORT_00_PIN_12 /* Onboard LED pins */#define FS_PARTITION_NAME "filesystem"voidhal_entry(void){ int result = RT_EOK; rt_kprintf("\nHello RT-Thread core0!\n"); structrt_device *mtd_dev = RT_NULL; externintfal_init(void); fal_init(); mtd_dev = fal_mtd_nor_device_create(FS_PARTITION_NAME); if (!mtd_dev) { LOG_E("Can't create a mtd device on '%s' partition.", FS_PARTITION_NAME); } else { if (dfs_mount(FS_PARTITION_NAME, "/", "lfs", 0, 0) == 0) { LOG_I("Filesystem initialized!"); if (mkdir("/webnet", 0777) == 0) { LOG_I("Created '/webnet' directory."); } } else { dfs_mkfs("lfs", FS_PARTITION_NAME); if (dfs_mount("filesystem", "/", "lfs", 0, 0) == 0) { LOG_I("Filesystem initialized!"); } else { LOG_E("Failed to initialize filesystem!"); } } } /* Initialize WiFi device and auto-connect */ LOG_I("Starting WiFi auto-connection..."); LOG_I("Connecting to SSID: %s", BSP_WIFI_SSID); /* Wait for WiFi device to be ready */ rt_thread_mdelay(2000); /* Connect to WiFi network */ result = rt_wlan_connect(BSP_WIFI_SSID, BSP_WIFI_PASSWORD); if (result == RT_EOK) { LOG_I("WiFi connection initiated successfully!"); /* Wait for connection to complete */ rt_thread_mdelay(5000); /* Check connection status */ if (rt_wlan_is_connected()) { LOG_I("WiFi connected successfully!"); /* Get and print IP address */ structrt_wlan_infoinfo; if (rt_wlan_get_info(&info) == RT_EOK) { LOG_I("Connected to: %s", info.ssid.val); LOG_I("RSSI: %d dBm", info.rssi); } } else { LOG_E("WiFi connection failed!"); } } else { LOG_E("Failed to initiate WiFi connection: %d", result); } while (1) { rt_pin_write(LED_PIN_0, PIN_HIGH); rt_thread_mdelay(1000); rt_pin_write(LED_PIN_0, PIN_LOW); rt_thread_mdelay(1000); }}

5AIOT大模型对话实验

下面的实验需要基于 Wi-Fi 工程基础上进行相关配置

使能 llmchat 软件包

点击添加软件包按钮，搜索并添加 llmchat软件包：

添加成功后如下图

通义千问账号注册

1.注册阿里云账号，并进入阿里云百炼平台：选择模型广场→API调用示例

2.模型使用需要充值，请酌情充值即可，这里演示充值1元就可使用：

https://billing-cost.console.aliyun.com/fortune/fund-management/recharge

3.接下来新建大模型的 API-KEY：

4.填写描述后即可创建API-KEY（SK-XXXXXXXX）:

5.打开 RT-Thread Settings，进入 RT-Thread online packages → AI packages → Large Language Models(LLM) for RT-Thread 目录下；按照下图进行配置：

llm qwen user api key：API KEY 需要在通义千问控制台获取；
llm qwen model name：选择模型的名称默认是 qwen-plus；

6.按照下图修改llm配置，同时使能 llmchat webnet 模式：

注意：如果在配置中没有显示“Enable llmchat webnet mode”选项，请进入到下图所示的目录下，编辑Kconfig文件：

请按照下图所示，添加相关代码，注意缩进：

config PKG_LLMCHAT_WEBNET_MODE bool"Enable llmchat webnet mode" select PKG_USING_WEBNET select WEBNET_USING_CGI default n help After using the WebNet model, you can communicate with the llm using a browser webpage.

保存，回到 Studio 中重新打开 RT-Thread Settings 即可显示此选项

7.进入 RT-Thread online packages → IoT - internet of things → WebClient: A HTTP/HTTPS Client for RT-Thread 选择MbedTLS support

8.进入 RT-Thread online packages → security packages → mbedtls 菜单，修改 Maxium fragment length in bytes 字段为 6144（否则TLS会握手失败）

9.使能旧版本兼容支持：

10.修改 LWIP 配置，增加 LWIP 线程栈大小：

11.使能组件-驱动中的软件模拟RTC设备：

12.退出保存配置，studio会自动拉取软件包：

13.修改 webnet线程栈大小为 8K：

14.软件包下载成功后，会在本地 packages下显示：

15.编译，烧录工程

6 效果演示

7 获取同款硬件