diff --git a/agent/proto.py b/agent/proto.py
index c59087d..8598169 100644
--- a/agent/proto.py
+++ b/agent/proto.py
@@ -109,7 +109,7 @@ def compute_intr_reward(self, obs, step):
         # enqueue candidates
         ptr = self.queue_ptr
         self.queue[ptr:ptr + self.num_protos] = z[candidates]
-        queue_ptr = (ptr + self.num_protos) % self.queue.shape[0]
+        self.queue_ptr = (ptr + self.num_protos) % self.queue.shape[0]
 
         # compute distances between the batch and the queue of candidates
         z_to_q = torch.norm(z[:, None, :] - self.queue[None, :, :], dim=2, p=2)