diff --git a/map.h b/map.h
index 6657ae8..ba8645f 100644
--- a/map.h
+++ b/map.h
@@ -81,7 +81,7 @@
 #define LCR_BLOCK_RAMP_12             0x07 ///< plain ramp, 1/2 size
 #define LCR_BLOCK_RAMP_14             0x08 ///< plain ramp, 1/4 size
 #define LCR_BLOCK_RAMP_CURVED_PLAT    0x09 ///< curved ramp with top platgform
-#define LCR_BLOCK_RAMP_CURVED   0x0a ///< curv. ramp without top platf.
+#define LCR_BLOCK_RAMP_CURVED         0x0a ///< curv. ramp without top platf.
 #define LCR_BLOCK_RAMP_CURVED_WALL    0x0b ///< curved ramp plus small wall
 #define LCR_BLOCK_RAMP_STEEP          0x0c ///< extremely steep ramp
 #define LCR_BLOCK_CORNER              0x0d ///< diagonal corner
@@ -112,6 +112,17 @@
                                              but makes a hollow one */
 #define LCR_BLOCK_START               0x83 ///< specifies start block position
 
+#define LCR_MAP_BLOCK_CACHE_SIZE (8 * 2)   /// do not change
+
+/**
+  Cache for accelerating LCR_mapGetBlockAtFast, consists of 8 2-item records,
+  the first record item stores block coord number, the second one stores the
+  value returned by LCR_mapGetBlockAtFast (-1 is 0xffffffff). The record index
+  depends on the block coords: lowest bit is x % 2, middle bit y % 2, highest
+  one z % 2.
+*/
+uint32_t _LCR_mapBlockCache[LCR_MAP_BLOCK_CACHE_SIZE];
+
 struct
 {
   uint16_t blockCount;
@@ -390,7 +401,12 @@ uint8_t LCR_mapLoad(const uint8_t *map)
     
     map += LCR_BLOCK_SIZE;
   }
-  
+
+  LCR_LOG2("clearing map block cache")
+
+  for (int i = 0; i < LCR_MAP_BLOCK_CACHE_SIZE; ++i)
+    _LCR_mapBlockCache[i] = 0xffffffff;
+
   LCR_LOG2("map loaded")
 
   return 1;
@@ -406,6 +422,14 @@ int LCR_mapGetBlockAtFast(uint8_t x, uint8_t y, uint8_t z,
   // binary search (the blocks are sorted)
 
   uint32_t n = LCR_mapBlockCoordsToCoordNumber(x,y,z);
+  uint8_t cacheIndex = 2 * ((x % 2) | ((y % 2) << 1) | ((z % 2) << 2));
+
+  if (_LCR_mapBlockCache[cacheIndex] == n)
+    return
+      (_LCR_mapBlockCache[cacheIndex + 1] != 0xffffffff) ?
+        ((int) _LCR_mapBlockCache[cacheIndex + 1]) : -1;
+
+  _LCR_mapBlockCache[cacheIndex] = n;
 
   while (start <= end)
   {
@@ -419,9 +443,13 @@ int LCR_mapGetBlockAtFast(uint8_t x, uint8_t y, uint8_t z,
     else if (n2 > n)
       end = m - 1;
     else
+    {
+      _LCR_mapBlockCache[cacheIndex + 1] = m;
       return m;
+    }
   }
 
+  _LCR_mapBlockCache[cacheIndex + 1] = 0xffffffff;
   return -1;
 }
 
diff --git a/racing.h b/racing.h
index 2d653f9..3fd02b5 100644
--- a/racing.h
+++ b/racing.h
@@ -9,10 +9,15 @@ typedef int32_t LCR_GameUnit;  ///< abstract game unit
 
 #define LCR_GAME_UNIT 1024     ///< length of map square in LCR_GameUnits
 
-#define LCR_RACING_INPUT_FORW  0x01
-#define LCR_RACING_INPUT_RIGHT 0x02
-#define LCR_RACING_INPUT_BACK  0x04
-#define LCR_RACING_INPUT_LEFT  0x08
+#define LCR_RACING_INPUT_FORW     0x01
+#define LCR_RACING_INPUT_RIGHT    0x02
+#define LCR_RACING_INPUT_BACK     0x04
+#define LCR_RACING_INPUT_LEFT     0x08
+
+#define LCR_RACING_EVENT_CP_TAKEN    0x0001
+#define LCR_RACING_EVENT_FINISHED    0x0002
+#define LCR_RACING_EVENT_CRASH_SMALL 0x0004
+#define LCR_RACING_EVENT_CRASH_BIG   0x0008
 
 #define LCR_PHYSICS_UNIT       2048 ///< length of map square for physics engine
   
@@ -571,11 +576,14 @@ int _LCR_racingCarShapeOK(void)
 
 /**
   Updates the racing physics world, call every LCR_RACING_TICK_MS milliseconds.
+  Returns a set of events (logically ORed) that occured during this step.
 */
-void LCR_racingStep(unsigned int input)
+uint32_t LCR_racingStep(unsigned int input)
 {
+printf("---------\n");
   LCR_LOG2("racing step start");
 
+  uint32_t result = 0;
   TPE_Vec3 carForw, carRight, carUp;
   uint8_t groundMat = LCR_BLOCK_MATERIAL_CONCRETE; // material under wheels
   int groundBlockIndex = -1;
@@ -870,6 +878,8 @@ void LCR_racingStep(unsigned int input)
   LCR_racing.tick++;
 
   LCR_LOG2("racing step end");
+
+  return result;
 }
 
 void LCR_physicsDebugDraw(LCR_GameUnit camPos[3], LCR_GameUnit camRot[2],
diff --git a/renderer.h b/renderer.h
index 1b25c7f..a257385 100644
--- a/renderer.h
+++ b/renderer.h
@@ -776,8 +776,6 @@ uint8_t _LCR_buildMapModel(void)
     }
   }
 
-  // TODO: also cull the triangles in the loop by some N steps
-
   _LCR_cullHiddenMapTris();
 
   LCR_LOG1("map model built");
@@ -806,6 +804,47 @@ void _LCR_rendererComputeLOD(void)
   }
 }
 
+void LCR_rendererMarkTakenCP(int x, int y, int z)
+{
+  for (int i = 0; i < LCR_renderer.mapModel.triangleCount; ++i)
+    if ((LCR_renderer.mapTriangleData[i] & 0x0f) == LCR_RENDERER_MAT_CP0)
+    {
+      S3L_Unit point[3];
+
+      point[0] = 0;
+      point[1] = 0;
+      point[2] = 0;
+
+      for (int j = 0; j < 2; ++j)
+        for (int k = 0; k < 3; ++k)
+          point[k] += LCR_renderer.mapModel.vertices[
+            3 * LCR_renderer.mapModel.triangles[3 * i + j] + k] +
+              (LCR_MAP_SIZE_BLOCKS / 2) *
+                (k == 1 ? LCR_RENDERER_UNIT / 2 : LCR_RENDERER_UNIT);
+
+      point[0] /= 2;
+      point[1] /= 2;
+      point[2] /= 2;
+
+      if (point[0] / LCR_RENDERER_UNIT == x &&
+        point[1] / (LCR_RENDERER_UNIT / 2) == y &&
+        point[2] / LCR_RENDERER_UNIT == z)
+        LCR_renderer.mapTriangleData[i] = (LCR_renderer.mapTriangleData[i]
+          & 0xf0) | LCR_RENDERER_MAT_CP1;
+    }
+}
+
+/**
+  Call to reset currently loaded map, i.e. mark all checkpoints as untaken etc.
+*/
+void LCR_rendererRestart(void)
+{
+  for (int i = 0; i < LCR_renderer.mapModel.triangleCount; ++i)
+    if ((LCR_renderer.mapTriangleData[i] & 0x0f) == LCR_RENDERER_MAT_CP1)
+      LCR_renderer.mapTriangleData[i] = (LCR_renderer.mapTriangleData[i] & 0xf0)
+        | LCR_RENDERER_MAT_CP0;
+}
+
 uint8_t LCR_rendererInit(void)
 {
   LCR_LOG0("initializing renderer");