Judder introduced by the driver

[0x1100]

Tracking had always been terrible compared to a native OSVR application.

The root cause of this problem is the fact that the update of the OSVR context is done in the runframe() method of the server driver. This method is called at an even lower frequency than the framerate so SteamVR tries to extrapolate much more that it should have to. @mdigkin's pull request #103 helps a lot because, without it, SteamVR cannot extrapolate the orientation at all. But it still doesn't fix the problem.

The context should instead be updated in a separate thread, as advised by Valve. This reduces the need for extrapolation to a minimum and, along with #103, completely fixes the judder problem.

[bagleman]

Are valve and Sensics/Razer aware of this problem?

[russell-taylor]

Sensics is now aware of it, and we're looking into it. Thanks!

[0x1100]

Getting it to work is pretty straightforward. We quickly patched it to release the AIO installer without having to wait a new build.

diff --git a/src/ServerDriver_OSVR.cpp b/src/ServerDriver_OSVR.cpp
index af5cb92..d97b014 100644
--- a/src/ServerDriver_OSVR.cpp
+++ b/src/ServerDriver_OSVR.cpp
@@ -42,6 +42,23 @@
 #include <cstring>                  // for std::strcmp
 #include <string>                   // for std::string

+#include <chrono>
+#include <thread>
+#include <atomic>
+
+namespace {
+static std::thread      gs_thread;
+static std::atomic<int> gs_thread_quit = 0;
+static std::atomic<int> gs_thread_ms_wait = 1;
+static void gs_thread_work(osvr::clientkit::ClientContext ctx) {
+    while (!gs_thread_quit.load()) {
+        ctx.update();
+        std::this_thread::sleep_for(std::chrono::milliseconds(gs_thread_ms_wait.load()));
+    }
+    gs_thread_quit = 0;
+}
+} // anonymous namespace
+
 vr::EVRInitError ServerDriver_OSVR::Init(vr::IDriverLog* driver_log, vr::IServerDriverHost* driver_host, const char* user_driver_config_dir, const char* driver_install_dir)
 {
     if (driver_log)
@@ -52,11 +69,15 @@ vr::EVRInitError ServerDriver_OSVR::Init(vr::IDriverLog* driver_log, vr::IServer
     trackedDevices_.emplace_back(std::make_unique<OSVRTrackedDevice>(*(context_.get()), driver_host));
     trackedDevices_.emplace_back(std::make_unique<OSVRTrackingReference>(*(context_.get()), driver_host));

+    gs_thread = std::thread(gs_thread_work, context_->get());
+
     return vr::VRInitError_None;
 }

 void ServerDriver_OSVR::Cleanup()
 {
+    gs_thread_quit.store(1);
+    gs_thread.join();
     trackedDevices_.clear();
     context_.reset();
 }
@@ -101,7 +122,10 @@ vr::ITrackedDeviceServerDriver* ServerDriver_OSVR::FindTrackedDeviceDriver(const

 void ServerDriver_OSVR::RunFrame()
 {
-    context_->update();
+    // In a real driver, this should happen from some pose tracking thread.
+    // The RunFrame interval is unspecified and can be very irregular if some other
+    // driver blocks it for some periodic task.
+    //context_->update();
 }

 bool ServerDriver_OSVR::ShouldBlockStandbyMode()
@@ -112,11 +136,13 @@ bool ServerDriver_OSVR::ShouldBlockStandbyMode()
 void ServerDriver_OSVR::EnterStandby()
 {
     // TODO
+    gs_thread_ms_wait.store(100);
 }

 void ServerDriver_OSVR::LeaveStandby()
 {
     // TODO
+    gs_thread_ms_wait.store(1);
 }

 std::string ServerDriver_OSVR::getDeviceId(vr::ITrackedDeviceServerDriver* device)

[godbyk]

I've incorporated a version of this patch into the update-v1.0.6 branch.

Thanks!