How to Diagnose and Resolve Common AWS DeepRacer Operational Issues - AWS DeepRacer

How to Diagnose and Resolve Common AWS DeepRacer Operational Issues

As you explore reinforcement learning with your AWS DeepRacer vehicle, the device may become non functional. The following troubleshooting topics help you diagnose the problems and resolve the issues.

Why Doesn't the Video Player on the Device Console Show the Video Stream from My Vehicle's Camera?

After logging into the AWS DeepRacer device console, you don't see any live video streamed from the camera mounted on the AWS DeepRacer vehicle in the video player in Device Controls. The following could cause this issue:

  • The camera might have a loose connection to the USB port. Unplug the camera module from the vehicle, replug it into the USB port, power off the device, and then power on the device to restart it.

  • The camera might be defective. Use a known working camera from another AWS DeepRacer vehicle, if available, to test whether this is the cause.

Why Doesn't My AWS DeepRacer Vehicle Move?

You powered on your AWS DeepRacer vehicle, but you can't make it to move. The following could cause this issue:

  • The vehicle's power bank is not turned on or the power bank is not connected to the vehicle. Make sure to connect the provided USB-C-to-USB C cable between the USB-C port on the power bank and the USB-C port on the vehicle chassis. Verify that the LED indicators light up, which indicates the charge levels of the power bank. If not, push the power button on the power bank, and then push the power button on the vehicle's chassis to boot up the device. The device is booted up when its tail lights light up.

  • If the power bank is on and the vehicle is booted up, but the vehicle does not move in either manual or autonomous driving mode, check if the vehicle's battery under the vehicle chassis is charged and turned on. If not, recharge the vehicle battery and then turn it on after the battery is fully charged.

  • The vehicle battery cable connectors are not fully plugged into the device driving module power cable connector. Make sure the cable connectors are tightly coupled.

  • The battery cables are defective. Test this battery on another working vehicle, if possible, to test whether this is the cause.

  • The power switch of the vehicle battery is not turned on. Turn on the power switch and make sure you hear two beeps followed by a long beep.

Why Don't I See the Latest Device Update? How Do I Get the Latest Update?

Why is my AWS DeepRacer vehicle's software outdated?

  • No automatic update is performed on the device lately. You may need to perform a manual update.

  • The vehicle is not connected to the Internet. Make sure the vehicle is connected to a Wi-Fi or Ethernet network with internet access.

Why Isn't My AWS DeepRacer Vehicle Connected to My Wi-Fi Network?

When I check the network status on the vehicle's OS, I don't see the AWS DeepRacer vehicle connected to any Wi-Fi network. This could happen because of the following issues:

  • No Wi-Fi has been configured for the AWS DeepRacer vehicle. Follow this setup instruction to set up the Wi-Fi network for your vehicle.

  • The vehicle is out of the active network signal range. Make sure to operate the vehicle within the chosen Wi-Fi network range.

  • The vehicle's pre-configured Wi-Fi network doesn't match the available Wi-Fi network. Follow the setup instruction to set up the Wi-Fi network that does not require active CAPTCHA.

Why Does the AWS DeepRacer Device Console Page Take a Long Time to Load?

When I tried to open the device console of my AWS DeepRacer vehicle, the device console page appears to take a long time to load.

  • Your vehicle is down or off. Make sure the vehicle is powered on when the tail lights are on.

  • The IP address of you vehicle has been changed, most likely by your network's DHCP server. To find out the vehicle's new IP address, follow these setup instructions to sign in to the device console with the USB-USB cable connection between your computer and the vehicle. View the new IP address in Settings. Alternatively, you can examine the list of devices attached to your network to discover the new IP address. If you're not a network administrator, ask the administrator to investigate this for you.

Why Does a Model Fail to Perform Well When Deployed to an AWS DeepRacer Vehicle?

After training a model and deploying its artifacts to your AWS DeepRacer vehicle, sometimes the vehicle doesn't perform as expected. What went wrong?

In general, optimizing a trained model for transfer to a physical AWS DeepRacer vehicle is a challenging learning process. It often requires iterations through trial and error. For general guidelines on best practices, see Optimize Training AWS DeepRacer Models for Real Environments.

The following are some likely common factors affecting the model performance in your AWS DeepRacer vehicle:

  • Your model has not converged in training. Clone the model to continue the training or retrain the model for a longer period of time. Make sure the agent continuously finishes laps in the simulation (that is, 100% process towards the end of the training).

  • Your model was over-trained (that is, over-fitted). It fits too well to the training data, but doesn't generalize to unknown situations. Retrain the model with a more flexible or accommodating reward function or increase the granularities of the action space. You should also evaluate a trained model on different tracks to see if the model generalizes well.

  • Your AWS DeepRacer vehicle may not have been calibrated properly. To test whether this is true, switch to manual driving and see if the vehicle drives as expected. If it doesn't, calibrate the vehicle.

  • You are running the vehicle autonomously on a track that doesn't meet the requirements. For track requirements, see Build Your Physical Track for AWS DeepRacer.

  • There are too many objects close to the physical track, making the track significantly different from the simulated environment. Clear the track surroundings to make the physical track as close to the simulated one as possible.

  • Reflection from the track surface or a near-by object can create glare to confuse the camera. Adjust lighting and avoid making the track on smooth-surfaced concrete floors or with other shiny materials.