EmbodiedAI
/
rl_sar
mirror of https://github.com/fan-ziqi/rl_sar.git
3
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

feat: add cyberdog

This commit is contained in:
fan-ziqi 2024-03-29 19:36:58 +08:00
parent 5dd3492eb2
commit 41b4672d75
5 changed files with 542 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

50
README_CN.md
View File

@ -78,6 +78,8 @@ rosrun teleop_twist_keyboard teleop_twist_keyboard.py
### 实物
#### Unitree A1
与实物的连接可分为无线与有线形式
* 无线：连接机器人发出的Unitree开头的WIFI **（注意：无线连接可能会出现丢包断联甚至失控，请注意安全）**
@ -92,6 +94,54 @@ rosrun rl_sar rl_real
按下遥控器的**R2**键让机器人切换到默认站起姿态，按下**R1**键切换到RL控制模式，任意状态按下**L2**切换到最初的趴下姿态。左摇杆上下控制x左右控制yaw，右摇杆左右控制y。
#### Cyberdog1
1. 连接机器人
将本地PC连接至铁蛋的USB download type-c 接口(位于中间)，等待出现”L4T-README” 弹窗
```bash
ping 192.168.55.100 #本地PC被分配的ip
ssh mi@192.168.55.1 #登录nx应用板 ,密码123
athena_version -v #核对当前版本>=1.0.0.94
ssh root@192.168.55.233 #登录运动控制板
```
2. 进入电机控制模式
修改配置开关，激活用户控制模式，运行用户自己的控制器：
```bash
ssh root@192.168.55.233 #登录运动控制板
cd /robot
./initialize.sh #拷贝出厂代码到可读写的开发区（/mnt/UDISK/robot-software），切换到开发者模式，仅需执行一次
vi /mnt/UDISK/robot-software/config/user_code_ctrl_mode.txt #切换mode:1(0:默认模式，1用户代码控制电机模式),重启机器人生效
```
3. 用户电脑侧部署
运行在用户pc侧(linux)难以保证实时lcm通信，仅推荐编译验证和简单的位控测试
```bash
ping 192.168.55.233 #通过type c线连接Cyberdog的Download接口后，确认通信正常
ifconfig | grep -B 1 192.168.55.100 | grep "flags"| cut -d ':' -f1 #获取该ip对应网络设备，一般为usb0
sudo ifconfig usb0 multicast #usb0替换为上文获取的168.55.100对应网络设备,并配为多播
sudo route add -net 224.0.0.0 netmask 240.0.0.0 dev usb0 #添加路由表，usb0对应替换
```
启动控制程序
```bash
source devel/setup.bash
rosrun rl_sar rl_real
```
按下键盘上的**0**键让机器人切换到默认站起姿态，按下**P**键切换到RL控制模式，任意状态按下**1**键切换到最初的趴下姿态。WS控制x，AD控制yaw，JL控制y。
注：lcm通信若不成功，无法正常激活电机控制模式，log提示：Motor control mode has not been activated successfully
## 引用
如果您使用此代码或其部分内容，请引用以下内容：

8
src/rl_sar/CMakeLists.txt
View File

@ -68,4 +68,12 @@ add_executable(rl_real src/rl_real.cpp )
target_link_libraries(rl_real
${catkin_LIBRARIES} ${EXTRA_LIBS} "${TORCH_LIBRARIES}"
rl observation_buffer
)
add_subdirectory(library/cyberdog_motor_sdk)
include_directories(library/cyberdog_motor_sdk/include)
add_executable(rl_real_cyberdog src/rl_real_cyberdog.cpp )
target_link_libraries(rl_real_cyberdog
${catkin_LIBRARIES} ${EXTRA_LIBS} "${TORCH_LIBRARIES}"
rl observation_buffer cyberdog_motor_sdk
)

89
src/rl_sar/include/rl_real_cyberdog.hpp Normal file
View File

@ -0,0 +1,89 @@
#ifndef RL_REAL_HPP
#define RL_REAL_HPP
#include "../library/rl/rl.hpp"
#include "../library/observation_buffer/observation_buffer.hpp"
#include "unitree_legged_sdk/loop.h"
#include <boost/bind.hpp>
#include <CustomInterface.h>
#include <csignal>
// #include <signal.h>
#include <termios.h>
#include <sys/ioctl.h>
using namespace UNITREE_LEGGED_SDK;
using CyberdogData = Robot_Data;
using CyberdogCmd = Motor_Cmd;
enum RobotState {
STATE_WAITING = 0,
STATE_POS_GETUP,
STATE_RL_INIT,
STATE_RL_RUNNING,
STATE_POS_GETDOWN,
};
struct KeyBoard
{
RobotState robot_state;
float x = 0;
float y = 0;
float yaw = 0;
};
class RL_Real : public RL, public CustomInterface
{
public:
RL_Real();
~RL_Real();
void RunModel();
torch::Tensor Forward() override;
torch::Tensor ComputeObservation() override;
ObservationBuffer history_obs_buf;
torch::Tensor history_obs;
int motiontime = 0;
CyberdogData cyberdogData;
CyberdogCmd cyberdogCmd;
void UserCode();
long long count = 0;
void RobotControl();
std::shared_ptr<LoopFunc> loop_control;
std::shared_ptr<LoopFunc> loop_rl;
std::shared_ptr<LoopFunc> loop_plot;
float getup_percent = 0.0;
float getdown_percent = 0.0;
float start_pos[12];
float now_pos[12];
int robot_state = STATE_WAITING;
std::vector<double> plot_t;
std::vector<std::vector<double>> plot_real_joint_pos, plot_target_joint_pos;
void Plot();
void run_keyboard();
KeyBoard keyboard;
std::thread _keyboardThread;
private:
std::vector<std::string> joint_names;
std::vector<double> joint_positions;
std::vector<double> joint_velocities;
int dof_mapping[12] = {3, 4, 5, 0, 1, 2, 9, 10, 11, 6, 7, 8};
int hip_scale_reduction_indices[4] = {0, 3, 6, 9};
std::chrono::high_resolution_clock::time_point start_time;
// other rl module
torch::jit::script::Module encoder;
torch::jit::script::Module vq;
};
#endif

6
src/rl_sar/src/rl_real.cpp
View File

@ -10,9 +10,9 @@ RL_Real::RL_Real() : safe(LeggedType::A1), udp(LOWLEVEL)
start_time = std::chrono::high_resolution_clock::now();
std::string actor_path = std::string(CMAKE_CURRENT_SOURCE_DIR) + "/models/actor.pt";
std::string encoder_path = std::string(CMAKE_CURRENT_SOURCE_DIR) + "/models/encoder.pt";
std::string vq_path = std::string(CMAKE_CURRENT_SOURCE_DIR) + "/models/vq_layer.pt";
std::string actor_path = std::string(CMAKE_CURRENT_SOURCE_DIR) + "/models/base/actor.pt";
std::string encoder_path = std::string(CMAKE_CURRENT_SOURCE_DIR) + "/models/base/encoder.pt";
std::string vq_path = std::string(CMAKE_CURRENT_SOURCE_DIR) + "/models/base/vq_layer.pt";
this->actor = torch::jit::load(actor_path);
this->encoder = torch::jit::load(encoder_path);

392
src/rl_sar/src/rl_real_cyberdog.cpp Normal file
View File

@ -0,0 +1,392 @@
#include "../include/rl_real_cyberdog.hpp"
// #define PLOT
RL_Real rl_sar;
RL_Real::RL_Real() : CustomInterface(500)
{
start_time = std::chrono::high_resolution_clock::now();
std::string actor_path = std::string(CMAKE_CURRENT_SOURCE_DIR) + "/models/cyberdog/actor.pt";
std::string encoder_path = std::string(CMAKE_CURRENT_SOURCE_DIR) + "/models/cyberdog/encoder.pt";
std::string vq_path = std::string(CMAKE_CURRENT_SOURCE_DIR) + "/models/cyberdog/vq_layer.pt";
this->actor = torch::jit::load(actor_path);
this->encoder = torch::jit::load(encoder_path);
this->vq = torch::jit::load(vq_path);
this->InitObservations();
this->params.num_observations = 45;
this->params.clip_obs = 100.0;
this->params.clip_actions = 100.0;
this->params.damping = 0.5;
this->params.stiffness = 20;
this->params.d_gains = torch::ones(12) * this->params.damping;
this->params.p_gains = torch::ones(12) * this->params.stiffness;
this->params.action_scale = 0.25;
this->params.hip_scale_reduction = 0.5;
this->params.num_of_dofs = 12;
this->params.lin_vel_scale = 2.0;
this->params.ang_vel_scale = 0.25;
this->params.dof_pos_scale = 1.0;
this->params.dof_vel_scale = 0.05;
this->params.commands_scale = torch::tensor({this->params.lin_vel_scale, this->params.lin_vel_scale, this->params.ang_vel_scale});
this->params.torque_limits = torch::tensor({{20.0, 55.0, 55.0,
20.0, 55.0, 55.0,
20.0, 55.0, 55.0,
20.0, 55.0, 55.0}});
// hip, thigh, calf
this->params.default_dof_pos = torch::tensor({{ 0.1000, 0.8000, -1.5000, // FL
-0.1000, 0.8000, -1.5000, // FR
0.1000, 1.0000, -1.5000, // RR
-0.1000, 1.0000, -1.5000}});// RL
this->history_obs_buf = ObservationBuffer(1, this->params.num_observations, 6);
output_torques = torch::tensor({{0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0}});
output_dof_pos = params.default_dof_pos;
plot_real_joint_pos.resize(12);
plot_target_joint_pos.resize(12);
loop_control = std::make_shared<LoopFunc>("loop_control", 0.002, boost::bind(&RL_Real::RobotControl, this));
loop_rl = std::make_shared<LoopFunc>("loop_rl" , 0.02 , boost::bind(&RL_Real::RunModel, this));
loop_control->start();
#ifdef PLOT
loop_plot = std::make_shared<LoopFunc>("loop_plot" , 0.002, boost::bind(&RL_Real::Plot, this));
loop_plot->start();
#endif
_keyboardThread = std::thread(&RL_Real::run_keyboard, this);
}
RL_Real::~RL_Real()
{
loop_control->shutdown();
loop_rl->shutdown();
#ifdef PLOT
loop_plot->shutdown();
#endif
printf("exit\n");
}
void RL_Real::RobotControl()
{
std::cout << "robot_state" << keyboard.robot_state
<< " x" << keyboard.x << " y" << keyboard.y << " yaw" << keyboard.yaw
<< "\r";
motiontime++;
// waiting
if(robot_state == STATE_WAITING)
{
for(int i = 0; i < 12; ++i)
{
cyberdogCmd.q_des[i] = cyberdogData.q[i];
}
if(keyboard.robot_state == STATE_POS_GETUP)
{
keyboard.robot_state = STATE_WAITING;
getup_percent = 0.0;
for(int i = 0; i < 12; ++i)
{
now_pos[i] = cyberdogData.q[i];
start_pos[i] = now_pos[i];
}
robot_state = STATE_POS_GETUP;
}
}
// stand up (position control)
else if(robot_state == STATE_POS_GETUP)
{
if(getup_percent != 1)
{
getup_percent += 1 / 1000.0;
getup_percent = getup_percent > 1 ? 1 : getup_percent;
for(int i = 0; i < 12; ++i)
{
cyberdogCmd.q_des[i] = (1 - getup_percent) * now_pos[i] + getup_percent * params.default_dof_pos[0][dof_mapping[i]].item<double>();
cyberdogCmd.qd_des[i] = 0;
cyberdogCmd.kp_des[i] = 50;
cyberdogCmd.kd_des[i] = 3;
cyberdogCmd.tau_des[i] = 0;
}
printf("getting up %.3f%%\r", getup_percent*100.0);
}
if(keyboard.robot_state == STATE_RL_INIT)
{
keyboard.robot_state = STATE_WAITING;
robot_state = STATE_RL_INIT;
}
else if(keyboard.robot_state == STATE_POS_GETDOWN)
{
keyboard.robot_state = STATE_WAITING;
getdown_percent = 0.0;
for(int i = 0; i < 12; ++i)
{
now_pos[i] = cyberdogData.q[i];
}
robot_state = STATE_POS_GETDOWN;
}
}
// init obs and start rl loop
else if(robot_state == STATE_RL_INIT)
{
if(getup_percent == 1)
{
robot_state = STATE_RL_RUNNING;
this->InitObservations();
printf("\nstart rl loop\n");
loop_rl->start();
}
}
// rl loop
else if(robot_state == STATE_RL_RUNNING)
{
for(int i = 0; i < 12; ++i)
{
// cyberdogCmd.q_des[i] = 0;
cyberdogCmd.q_des[i] = output_dof_pos[0][dof_mapping[i]].item<double>();
cyberdogCmd.qd_des[i] = 0;
cyberdogCmd.kp_des[i] = params.stiffness;
cyberdogCmd.kd_des[i] = params.damping;
// cyberdogCmd.tau_des[i] = output_torques[0][dof_mapping[i]].item<double>();
cyberdogCmd.tau_des[i] = 0;
}
if(keyboard.robot_state == STATE_POS_GETDOWN)
{
keyboard.robot_state = STATE_WAITING;
getdown_percent = 0.0;
for(int i = 0; i < 12; ++i)
{
now_pos[i] = cyberdogData.q[i];
}
robot_state = STATE_POS_GETDOWN;
}
}
// get down (position control)
else if(robot_state == STATE_POS_GETDOWN)
{
if(getdown_percent != 1)
{
getdown_percent += 1 / 1000.0;
getdown_percent = getdown_percent > 1 ? 1 : getdown_percent;
for(int i = 0; i < 12; ++i)
{
cyberdogCmd.q_des[i] = (1 - getdown_percent) * now_pos[i] + getdown_percent * start_pos[i];
cyberdogCmd.qd_des[i] = 0;
cyberdogCmd.kp_des[i] = 50;
cyberdogCmd.kd_des[i] = 3;
cyberdogCmd.tau_des[i] = 0;
}
printf("getting down %.3f%%\r", getdown_percent*100.0);
}
if(getdown_percent == 1)
{
robot_state = STATE_WAITING;
this->InitObservations();
printf("\nstop rl loop\n");
loop_rl->shutdown();
}
}
}
void RL_Real::UserCode()
{
cyberdogData = robot_data;
motor_cmd = cyberdogCmd;
}
static bool kbhit()
{
termios term;
tcgetattr(0, &term);
termios term2 = term;
term2.c_lflag &= ~ICANON;
tcsetattr(0, TCSANOW, &term2);
int byteswaiting;
ioctl(0, FIONREAD, &byteswaiting);
tcsetattr(0, TCSANOW, &term);
return byteswaiting > 0;
}
void RL_Real::run_keyboard()
{
int c;
// Check for keyboard input
while(true)
{
if(kbhit())
{
c = fgetc(stdin);
switch(c)
{
case '0':
keyboard.robot_state = STATE_POS_GETUP;
break;
case 'p':
keyboard.robot_state = STATE_RL_INIT;
break;
case '1':
keyboard.robot_state = STATE_POS_GETDOWN;
break;
case 'q':
break;
case 'w':
keyboard.x += 0.1;
break;
case 's':
keyboard.x -= 0.1;
break;
case 'a':
keyboard.yaw += 0.1;
break;
case 'd':
keyboard.yaw -= 0.1;
break;
case 'i':
break;
case 'k':
break;
case 'j':
keyboard.y += 0.1;
break;
case 'l':
keyboard.y -= 0.1;
break;
case ' ':
keyboard.x = 0;
keyboard.y = 0;
keyboard.yaw = 0;
break;
default:
break;
}
}
usleep(10000);
}
}
void RL_Real::RunModel()
{
if(robot_state == STATE_RL_RUNNING)
{
// auto duration = std::chrono::duration_cast<std::chrono::microseconds>(std::chrono::high_resolution_clock::now() - start_time).count();
// std::cout << "Execution time: " << duration << " microseconds" << std::endl;
// start_time = std::chrono::high_resolution_clock::now();
// printf("%f, %f, %f\n",
// cyberdogData.omega[0], cyberdogData.omega[1], cyberdogData.omega[2]);
// printf("%f, %f, %f, %f\n",
// cyberdogData.quat[1], cyberdogData.quat[2], cyberdogData.quat[3], cyberdogData.quat[0]);
// printf("%f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f\n",
// cyberdogData.q[3], cyberdogData.q[4], cyberdogData.q[5],
// cyberdogData.q[0], cyberdogData.q[1], cyberdogData.q[2],
// cyberdogData.q[9], cyberdogData.q[10], cyberdogData.q[11],
// cyberdogData.q[6], cyberdogData.q[7], cyberdogData.q[8]);
// printf("%f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f\n",
// cyberdogData.qd[3], cyberdogData.qd[4], cyberdogData.qd[5],
// cyberdogData.qd[0], cyberdogData.qd[1], cyberdogData.qd[2],
// cyberdogData.qd[9], cyberdogData.qd[10], cyberdogData.qd[11],
// cyberdogData.qd[6], cyberdogData.qd[7], cyberdogData.qd[8]);
this->obs.ang_vel = torch::tensor({{cyberdogData.omega[0], cyberdogData.omega[1], cyberdogData.omega[2]}});
this->obs.commands = torch::tensor({{keyboard.x, keyboard.y, keyboard.yaw}});
this->obs.base_quat = torch::tensor({{cyberdogData.quat[1], cyberdogData.quat[2], cyberdogData.quat[3], cyberdogData.quat[0]}});
this->obs.dof_pos = torch::tensor({{cyberdogData.q[3], cyberdogData.q[4], cyberdogData.q[5],
cyberdogData.q[0], cyberdogData.q[1], cyberdogData.q[2],
cyberdogData.q[9], cyberdogData.q[10], cyberdogData.q[11],
cyberdogData.q[6], cyberdogData.q[7], cyberdogData.q[8]}});
this->obs.dof_vel = torch::tensor({{cyberdogData.qd[3], cyberdogData.qd[4], cyberdogData.qd[5],
cyberdogData.qd[0], cyberdogData.qd[1], cyberdogData.qd[2],
cyberdogData.qd[9], cyberdogData.qd[10], cyberdogData.qd[11],
cyberdogData.qd[6], cyberdogData.qd[7], cyberdogData.qd[8]}});
torch::Tensor actions = this->Forward();
for (int i : hip_scale_reduction_indices)
{
actions[0][i] *= this->params.hip_scale_reduction;
}
output_torques = this->ComputeTorques(actions);
output_dof_pos = this->ComputePosition(actions);
}
}
torch::Tensor RL_Real::ComputeObservation()
{
torch::Tensor obs = torch::cat({// this->QuatRotateInverse(this->obs.base_quat, this->obs.lin_vel) * this->params.lin_vel_scale,
this->QuatRotateInverse(this->obs.base_quat, this->obs.ang_vel) * this->params.ang_vel_scale,
this->QuatRotateInverse(this->obs.base_quat, this->obs.gravity_vec),
this->obs.commands * this->params.commands_scale,
(this->obs.dof_pos - this->params.default_dof_pos) * this->params.dof_pos_scale,
this->obs.dof_vel * this->params.dof_vel_scale,
this->obs.actions
},1);
obs = torch::clamp(obs, -this->params.clip_obs, this->params.clip_obs);
return obs;
}
torch::Tensor RL_Real::Forward()
{
torch::Tensor obs = this->ComputeObservation();
history_obs_buf.insert(obs);
history_obs = history_obs_buf.get_obs_vec({0, 1, 2, 3, 4, 5});
torch::Tensor encoding = this->encoder.forward({history_obs}).toTensor();
torch::Tensor z = this->vq.forward({encoding}).toTensor();
torch::Tensor actor_input = torch::cat({obs, z}, 1);
torch::Tensor action = this->actor.forward({actor_input}).toTensor();
this->obs.actions = action;
torch::Tensor clamped = torch::clamp(action, -this->params.clip_actions, this->params.clip_actions);
return clamped;
}
void RL_Real::Plot()
{
plot_t.push_back(motiontime);
plt::cla();
plt::clf();
for(int i = 0; i < 12; ++i)
{
plot_real_joint_pos[i].push_back(cyberdogData.q[i]);
plot_target_joint_pos[i].push_back(cyberdogCmd.q_des[i]);
plt::subplot(4, 3, i+1);
plt::named_plot("_real_joint_pos", plot_t, plot_real_joint_pos[i], "r");
plt::named_plot("_target_joint_pos", plot_t, plot_target_joint_pos[i], "b");
plt::xlim(motiontime-10000, motiontime);
}
// plt::legend();
plt::pause(0.0001);
}
void signalHandler(int signum)
{
exit(0);
}
int main(int argc, char **argv)
{
signal(SIGINT, signalHandler);
while(1)
{
sleep(10);
};
return 0;
}