[Minor] typo in "Multi-Agent Reinforcement Learning (PPO) with TorchRL Tutorial" docs (#2938)

Author: sendhil

tutorials/sphinx-tutorials/multiagent_ppo.py

@@ -57,7 +57,7 @@
 #
 # This type of algorithms is usually trained *on-policy*. This means that, at every learning iteration, we have a
 # **sampling** and a **training** phase. In the **sampling** phase of iteration :math:`t`, rollouts are collected
-# form agents' interactions in the environment using the current policies :math:`\mathbf{\pi}_t`.
+# from agents' interactions in the environment using the current policies :math:`\mathbf{\pi}_t`.
 # In the **training** phase, all the collected rollouts are immediately fed to the training process to perform
 # backpropagation. This leads to updated policies which are then used again for sampling.
 # The execution of this process in a loop constitutes *on-policy learning*.