How to train a differential robot using reinforcement learning?

I need my robot to reach a specific point in a given area. Currently, the algorithm $($DDPG$)$ consists of the following parameters:

$-$ State: Robot's x and y position, its orientation, and the reading of an ultrasonic sensor$,$ for a total of $4$ parameters.

$-$ Goal: The goal is composed of the positions to be reached.

$-$ Actions: The actions generated are the linear and angular velocity of the robot.

$-$ Reward: It is given by the following function:

def reward$\_$function$($self$,$ objective$\_$distance, obstacle$\_$distance, a$\_$velocity, l$\_$velocity, time$\_$on$\_$objective$)$:

# Objective distance penalization

reward$\_$objective$\_$distance $=-1*($objective$\_$distance$)**2/(4\mathrm{.}0)**2$

# Obstacle distance penalization

reward$\_$obstacle$\_$distance $=(-1*((1/($obstacle$\_$distance$+0\mathrm{.}06)))/16)+0\mathrm{.}07$

# High speeds penalizations

reward$\_$velocity $=-1*($a$\_$velocity$**2+$ l$\_$velocity$**2)/8\mathrm{.}0$

# Reward for staying on target

reward$\_$time $=$ time$\_$on$\_$objective

reward $=$ reward$\_$objective$\_$distance $+$ reward$\_$obstacle$\_$distance $+\backslash$

reward$\_$velocity $+$ reward$\_$time

return reward

$-$ Neural network parameters:

$*$ Actor: Two hidden layers of $256$ neurons, learning rate of $0\mathrm{.}0001$

$*$ Critic: Two hidden layers of $256$ neurons, learning rate of $0\mathrm{.}001$

However, in all simulations I get very bad results.

Could you give me some tips to improve the training of the robot?