Improved Obstacle Handling in Tesla FSD Beta #2

Table of Contents

1. Introduction
2. testing the Beta in Obstacle Scenarios
   • Testing the Gray Starling Box
   • Testing the Stationary Box
   • Testing the Moving Box
   • Testing the Box in Motion
   • Testing the Wheel in Upright Position
   • Testing the Wheel on Its Side
   • Testing the Wheel in Motion
3. Wife Testing in Various Scenarios
   • Wife Test 1: Parking Obstacle
   • Wife Test 2: Lane Crossing
   • Wife Test 3: Path Planning with Advanced Path Blockage
   • Wife Test 4: Head-on Approach
   • Wife Test 5: Parked Lengthwise Across the Road
4. Conclusion

Introduction

Welcome to another episode of Full Self-Driving Torture! In this episode, we will be pushing the limits of the Beta version of the self-driving system. We'll be testing its performance in various obstacle scenarios and also conducting some unique wife tests to see how it reacts. So, let's dive in and see how the Beta fares!

Testing the Beta in Obstacle Scenarios

Testing the Gray Starling Box

In our previous obstacle testing video, we encountered some difficulties with the Beta controlling for the gray Starling box. We decided to start this episode by retesting the system's performance in this specific Scenario. Surprisingly, the Beta confidently maneuvers around the box without any hesitation this time. It seems like there have been some improvements in the latest version.

Testing the Stationary Box

To further assess the system's capabilities, we put the box in a more challenging spot on the road, blending it with the shadow of the fence. The Beta pleasantly surprises us again by successfully going around the box without any issues. The results are consistent, indicating that the system is adapting well to different stationary obstacles.

Testing the Moving Box

Next, we up the difficulty level by putting the box in motion. My wife, Carol, stands on the opposite side of the street and kicks the box directly into our path at the last moment. Although the Beta attempts to avoid the moving box, it doesn't go too far into the shoulder. Despite that, it manages to navigate past the obstacle without slowing down. This test demonstrates the system's ability to handle unexpected obstacles in its path.

Testing the Wheel in Upright Position

Moving on to a bigger and more destructive obstacle, we wanted to see if the Beta can predict and avoid a wheel in the upright position. In previous tests, version 10 struggled to handle this scenario. However, in version 11, we witnessed significant improvements. The Beta successfully avoids the wheel, showing enhanced comfort and smoother maneuvering.

Testing the Wheel on Its Side

To further challenge the system, we test the Beta's capability to avoid a wheel placed on its side. Although there is some indecisiveness with the path planner, the Beta manages to navigate around the wheel without much drama. The system shows improvement in handling this specific obstacle, although there were some variations in its path selection.

Testing the Wheel in Motion

The most challenging scenario for the Beta in version 10 was predicting and reacting to a wheel in motion. In this test, we Roll the wheel in front of the vehicle, and unfortunately, version 11 still struggles with this situation. The Beta slows down slightly after the wheel is rolled, but it doesn't anticipate the wheel's movement effectively. We hope future versions will overcome this limitation with advanced occupancy network improvements.

Wife Testing in Various Scenarios

Wife Test 1: Parking Obstacle

In response to viewer feedback, we decided to conduct some wife tests. In the first challenge, my wife, Carol, parks her car in the middle of the road, and the Beta needs to navigate around it. The system performs impressively, slowing down to an appropriate speed and folding the mirrors to pass through the narrow space. It's worth noting that this test was rather terrifying due to the close proximity between the vehicles.

Wife Test 2: Lane Crossing

To assess the Beta's reaction to a sudden lane crossing, my wife starts turning into my lane as we approach each other. Surprisingly, the Beta shows no significant reaction to this maneuver, even when my wife increases the speed. It seems that the advanced path planning, indicated by a blue visualization, doesn't activate unless the other vehicle fully crosses into our lane.

Wife Test 3: Path Planning with Advanced Path Blockage

Continuing the wife tests, we push the limits by engaging in a head-on approach with no stopping or disengagements. Although my wife's car is only moving at a slow speed, the Beta struggles to handle this situation. It attempts to yield to the right but then abruptly moves to the left before ultimately stopping when it misinterprets the position of the rear of the other vehicle. This behavior was unexpected and highlighted a lack of consistency in the system's decision-making process.

Wife Test 4: Head-on Approach

To further explore the system's response to a head-on approach, we repeat the test. This time, the Beta shows a similar pattern of hesitation and confusion, but it eventually stops when the visualization indicates that it is facing the rear of the other vehicle. This erratic behavior showcases the complexity involved in path planning and the need for further improvements.

Wife Test 5: Parked Lengthwise Across the Road

In the final wife test, my wife parks her car lengthwise across the road. Surprisingly, this time the Beta successfully recognizes the obstacle and displays a blue visualization, indicating advanced path planning. This outcome confirms that the system's performance can vary even in similar scenarios and highlights the need for continued development.

Conclusion

In this episode of Full Self-Driving Torture, we put the Beta version to the test in various obstacle scenarios and conducted unique wife tests. Overall, the Beta shows improvement in maneuvering around stationary and moving obstacles. However, it still struggles with predicting the movement of certain objects, indicating the need for further advancements in path planning and occupant network technology. Despite its limitations, it's important to remember that these tests are designed to stress the system and track progress over time. We remain hopeful that future versions will continue to enhance the system's capabilities, making autonomous driving safer and more reliable.

Highlights

• The Beta version shows improvement in maneuvering around stationary and moving obstacles.
• The system struggles with predicting the movement of certain objects, highlighting the need for further advancements in path planning.
• Wife tests reveal inconsistencies in the system's decision-making process during complex scenarios.
• The Beta's performance varies in similar scenarios, indicating the ongoing development required for advanced path planning.
• Stress testing allows for tracking progress and identifying areas for improvement in autonomous driving technology.

Frequently Asked Questions (FAQ)

Q: Is the Beta version of the self-driving system fully autonomous? A: No, the Beta version is still under development and requires driver supervision. It is important to remain vigilant and ready to take control if necessary.

Q: Are the obstacle scenarios shown in the video realistic? A: Some scenarios are designed to stress test the system and may not occur frequently in everyday driving. They aim to assess the system's performance in challenging situations.

Q: Can the Beta version handle all obstacles? A: While the Beta has shown improvements in maneuvering around stationary and certain moving obstacles, it still has limitations in predicting the movement of certain objects. Further advancements are needed.

Q: What is the purpose of the wife tests? A: The wife tests aim to explore the system's response to real-life scenarios involving other vehicles on the road. They provide insights into the system's decision-making process and identify areas for improvement.

Q: How do future versions aim to address the limitations observed in the Beta? A: Future versions are expected to incorporate advanced occupancy network improvements, showcased in Tesla's AI day, to improve the system's ability to predict and react to moving obstacles.

Q: How can I support the Channel and sponsor? A: You can support the channel by purchasing cameras and products from insta360 through the provided links in the description. This helps support the channel's content creation.