77ea67e58154f02617cda121847158973eb19700,cistar-dev/cistar/envs/lane_changing.py,SimpleLaneChangingAccelerationEnvironment,compute_reward,#SimpleLaneChangingAccelerationEnvironment#Any#Any#,63

Before Change

        
        See parent class
        
        return -np.linalg.norm(velocity - self.env_params["target_velocity"])

    def getState(self):
        
       See parent class

After Change

        See parent class
        
        // upper bound used to ensure the reward is always positive
        max_cost = np.array([self.env_params["target_velocity"]]*self.scenario.num_vehicles)
        max_cost = np.linalg.norm(max_cost)

        // cost associated with being away from target velocity
        // if the vehicle"s velocity is more than twice the target velocity, the cost does not become worse
        cost = velocity - self.env_params["target_velocity"]
        cost = np.linalg.norm(cost)

        ////////////////////////////////////////////////////////////////////////////////////////////
        if any(velocity < 0):
            print("------------------------------")
            print(velocity)
            print(np.array(self.rl_ids)[np.array(velocity) < 0])
            print("------------------------------")
        ////////////////////////////////////////////////////////////////////////////////////////////

        return max_cost - cost

        // return np.linalg.norm(np.array([0]*len(velocity)) - np.array([50]*len(velocity))) - \
        //     np.linalg.norm(velocity - self.env_params["target_velocity"])

Instances