QOL improvement when building an XML? Unity Style Navigation for the simulate viewer

[rohit-kumar-j]

Hi all,

This is mostly an ease of use modification made to mujoco's simulate(I don't think this would improve simulate).

I don't like taking my keys off the keyboard... Let me rephrase that... I'd like to have an option to not use the mouse and still navigate around.
So I've make unity style keybinds to move the camera around when I build my robot.

Unity like movement:
Shift + W/A/S/D — translate along camera view axes
Shift + Q / E — world up / down
Ctrl + Alt + W/A/S/D — pitch / yaw, pivoting on the camera position

Since this deviates from the UiEvent() pipeline (key releases don't surface through OnKey, so continuous motion needs polling), I had to modify the main render loop a bit:

A new ApplyCameraKeys(Simulate*) helper runs once per frame inside the mutex, alongside UiEvent. It uses one new pure-virtual on bool PlatformUIAdapter::IsKeyPressed(int key) const with a one line glfwGetKey override in GlfwAdapter

If this is okay I could open a PR, or maybe there is a better way to do this. Either ways, I'm just sharing this to find any other takers for such a feature.

Please share your thoughts 🙌

Diff

diff --git a/simulate/glfw_adapter.cc b/simulate/glfw_adapter.cc index 6bae8822..53da6c03 100644 --- a/simulate/glfw_adapter.cc +++ b/simulate/glfw_adapter.cc @@ -231,6 +231,10 @@ bool GlfwAdapter::IsShiftKeyPressed() const { Glfw().glfwGetKey(window_, GLFW_KEY_RIGHT_SHIFT) == GLFW_PRESS; }

+bool GlfwAdapter::IsKeyPressed(int key) const {

• return Glfw().glfwGetKey(window_, key) == GLFW_PRESS;
  +}

bool GlfwAdapter::IsMouseButtonDownEvent(int act) const {
return act == GLFW_PRESS;
}
diff --git a/simulate/glfw_adapter.h b/simulate/glfw_adapter.h
index 1490c333..e1a252c0 100644
--- a/simulate/glfw_adapter.h
+++ b/simulate/glfw_adapter.h
@@ -52,6 +52,7 @@ class GlfwAdapter : public PlatformUIAdapter {
bool IsAltKeyPressed() const override;
bool IsCtrlKeyPressed() const override;
bool IsShiftKeyPressed() const override;

• bool IsKeyPressed(int key) const override;

  bool IsMouseButtonDownEvent(int act) const override;
  bool IsKeyDownEvent(int act) const override;
  diff --git a/simulate/platform_ui_adapter.h b/simulate/platform_ui_adapter.h
  index 64b0211d..0a9c9c82 100644
  --- a/simulate/platform_ui_adapter.h
  +++ b/simulate/platform_ui_adapter.h
  @@ -64,6 +64,7 @@ class PlatformUIAdapter {
  virtual bool IsAltKeyPressed() const = 0;
  virtual bool IsCtrlKeyPressed() const = 0;
  virtual bool IsShiftKeyPressed() const = 0;

• virtual bool IsKeyPressed(int key) const = 0;

  virtual bool IsMouseButtonDownEvent(int act) const = 0;
  virtual bool IsKeyDownEvent(int act) const = 0;
  diff --git a/simulate/simulate.cc b/simulate/simulate.cc
  index 76e2b788..4c4b8f06 100644
  --- a/simulate/simulate.cc
  +++ b/simulate/simulate.cc
  @@ -1578,6 +1578,76 @@ void UiModify(mjUI* ui, mjuiState* state, mjrContext* con) {
  mjui_update(-1, -1, ui, state, con);
  }

+// Continuous Unity-style flycam, polled once per frame via the platform UI
+// Shift + W/A/S/D : forward / left / back / right along camera axes
+// Shift + Q / E : world-up / world-down
+// Ctrl+Alt + W/S/A/D : pitch up/down, yaw left/right (pivot at camera position)
+void ApplyCameraKeys(mj::Simulate* sim) {

• if (sim->cam.type != mjCAMERA_FREE || !sim->m_) {
• sim->cam_move_initialized_ = false;
• return;
• }
• const mj::PlatformUIAdapter& ui = *sim->platform_ui;
• auto now = mj::Simulate::Clock::now();
• if (!sim->cam_move_initialized_) {
• sim->cam_move_last_time_ = now;
• sim->cam_move_initialized_ = true;
• return;
• }
• double dt = Seconds(now - sim->cam_move_last_time_).count();
• sim->cam_move_last_time_ = now;
• if (dt > 0.1) dt = 0.1;
• const bool shift = ui.IsShiftKeyPressed();
• const bool ctrl = ui.IsCtrlKeyPressed();
• const bool alt = ui.IsAltKeyPressed();
• auto held = [&](int k) { return ui.IsKeyPressed(k); };
• // Shift+Alt+W/S/A/D: rotate the view, keeping camera position fixed
• // W = pitch up S = pitch down
• // A = yaw left D = yaw right
• if (shift && alt) {
• const mjtNum rate_deg = 60.0;
• mjtNum d_az = 0, d_el = 0;
• if (held('A')) d_az += rate_deg * dt;
• if (held('D')) d_az -= rate_deg * dt;
• if (held('W')) d_el += rate_deg * dt;
• if (held('S')) d_el -= rate_deg * dt;
• if (d_az == 0 && d_el == 0) return;
• const mjtNum el0 = sim->cam.elevation * mjPI / 180.0;
• const mjtNum a0 = sim->cam.azimuth * mjPI / 180.0;
• const mjtNum el1 = (sim->cam.elevation + d_el) * mjPI / 180.0;
• const mjtNum a1 = (sim->cam.azimuth + d_az) * mjPI / 180.0;
• // forward vectors (MuJoCo: cam_pos = lookat - dist * forward)
• const mjtNum f0[3] = { mju_cos(el0)*mju_cos(a0), mju_cos(el0)*mju_sin(a0), mju_sin(el0) };
• const mjtNum f1[3] = { mju_cos(el1)*mju_cos(a1), mju_cos(el1)*mju_sin(a1), mju_sin(el1) };
• const mjtNum dist = sim->cam.distance;
• for (int i = 0; i < 3; ++i) sim->cam.lookat[i] += dist * (f1[i] - f0[i]);
• sim->cam.azimuth += d_az;
• sim->cam.elevation += d_el;
• return;
• }
• if (!shift || ctrl || alt) return;
• const mjtNum s = 0.6 * sim->m_->stat.extent * dt;
• mjtNum fwd[3], up[3], right[3];
• mju_f2n(fwd, sim->scn.camera[0].forward, 3);
• mju_f2n(up, sim->scn.camera[0].up, 3);
• mju_cross(right, fwd, up);
• if (held('W')) for (int i=0;i<3;++i) sim->cam.lookat[i] += s*fwd[i];
• if (held('S')) for (int i=0;i<3;++i) sim->cam.lookat[i] -= s*fwd[i];
• if (held('D')) for (int i=0;i<3;++i) sim->cam.lookat[i] += s*right[i];
• if (held('A')) for (int i=0;i<3;++i) sim->cam.lookat[i] -= s*right[i];
• if (held('Q')) sim->cam.lookat[2] += s;
• if (held('E')) sim->cam.lookat[2] -= s;
  +}

// handle UI event
void UiEvent(mjuiState* state) {
mj::Simulate* sim = static_castmj::Simulate*(state->userdata);
@@ -3006,6 +3076,8 @@ void Simulate::RenderLoop() {
cond_upload_.notify_all();
}

•  ApplyCameraKeys(this);
  

•  // update scene, doing a full sync if in fully managed mode
   if (!is_passive_) {
     Sync();
  

diff --git a/simulate/simulate.h b/simulate/simulate.h
index 27727ee7..3c8a6551 100644
--- a/simulate/simulate.h
+++ b/simulate/simulate.h
@@ -255,6 +255,10 @@ class Simulate {
// rendering: need sync
int camera = 0;

• // Continuous WASDQE flycam state
• std::chrono::time_point cam_move_last_time_;
• bool cam_move_initialized_ = false;

// abstract visualization
mjvScene scn;
mjvCamera& cam;

Screen.Recording.2026-05-15.at.1.38.43.PM.mp4