Cost update in MPPIPlanner._compute_rollout_costs(perturbed_actions)

[Jerome-Cong]

Hi all,

Thanks for the nice work! I have a question regarding the _compute_rollout_costs() method in the MPPIPlanner class:

        # action perturbation cost
        if self.terminal_state_cost:
            c = self.terminal_state_cost(states, actions)
            cost_samples += c
        cost_total += cost_samples.mean(dim=0)
        
        if self.mppi_mode == 'halton-spline':
            self.noise = self._update_distribution(cost_horizon, actions)

        return cost_total, states, actions

When updating cost_total of K samples, why is the mean over all samples used?