Lab 10a - Jetbot self-driving
Robotics II
Poznan University of Technology, Institute of Robotics and Machine Intelligence
Laboratory 7: NVIDIA Jetbot self-driving
Back to the course table of contents
The aim of this task is to create an autonomous control NVIDIA Jetbot robot.
- Check the instruction and example notebooks related to NVIDIA Jetbot.
- The whole project instruction is included in the repository README file. Go to and enjoy.
RULES AND REGULATIONS - “JetBot Grand Prix – Artificial Intelligence Robotics Challenge”
1. Objectives
The objectives of the challange are:
- the practical application of artificial intelligence methods in robotic systems,
- the development of teamwork skills,
- the design and testing of autonomous systems.
The course is conducted using the NVIDIA AI IoT JetBot platform
2. Teams
Students work in teams of 4-5 people
Each team: chooses a name (e.g., F1 style) is responsible for one JetBot (“race car”)
The team is fully responsible for: - the code - the AI model - the robot’s configuration
3. Structure of competition
We will have 7 labs for competition.
Lab 1
Introduction to the robot: - camera - motors - Jetson Nano access to the environment (Jupyter / SSH)
Lab 2
Understanding robot control and the basics of movement
Work with:
- motor control
- moving forward / backward / turning
- basics of calibration
- writing simple control code (laptop / gamepad)
- testing the robot’s movement
RACE 1 – Time Trial (manual)
- manual control (gamepad / laptop)
- completing the track against the clock
POINTS CALCULATOR:
\(P = P_{max} ⋅ \frac{T_{team}}{T_{best}}\)
where:
\(P_{max} = 25\)
\(T_{team}\)- the time of a given team
\(T_{best}\)- best time in round
LAB 3,4 – Data Collection, Model Training and AI Control
Understanding AI input data and building the first AI model
Work with:
- camera operation
- image recording
- create their own training data
- model training
- inference (robot control)
LAB 5 First AI Race
Work with:
- dataset improvement
- model tuning
- testing
RACE 2 – AI Challenge (scored)
- autonomous robot
- on track
POINTS CALCULATOR:
\(P = P_{max} ⋅ \frac{T_{team}}{T_{best}}\)
where:
\(P_{max} = 25\)
\(T_{team}\)- the time of a given team
\(T_{best}\)- best time in round
LAB 6 - Iteration and Improvements
Work with:
- problem analysis
- overfitting
- poor data
- changes in lighting
LAB 7 – Final: JetBot Grand Prix
Presentation of the final solution
FINAL RACE
- full autonomy
- complex track
POINTS CALCULATOR:
\(P = P_{max} ⋅ \frac{T_{team}}{T_{best}}\)
where:
\(P_{max} = 25\)
\(T_{team}\)- the time of a given team
\(T_{best}\)- best time in round
DETAILED REGULATIONS
JetBot Grand Prix – Rules and Participation Guidelines
1. General Rules
1.1. Every student must belong to a project team
1.2. Teams consist of 3-5 members
1.3. Each team works with one robot (JetBot)
1.4. The team is responsible for the entire solution (code, model, configuration)
1.5. All tasks must be completed independently by the team
1.6. Use of documentation and educational materials is allowed
1.7. Copying solutions without understanding them is prohibited
2. Team Organization
2.1. Each team must choose a name (used for identification)
2.2. Team composition is fixed for the entire semester
2.3. Any changes require instructor approval
2.4. Each member must be able to explain the project
2.5. Lack of participation may result in individual grading
3. Hardware and Environment
3.1. All teams must use the provided robots
3.2. Hardware configuration must remain consistent with course standards
3.3. Unauthorized hardware modifications are prohibited
3.4. Software modifications (code, AI models) are allowed
3.5. Each team is responsible for the technical condition of the robot
3.6. Any hardware issues must be reported immediately
4. Artificial Intelligence Requirements
4.1. Autonomous tasks must use camera input
4.2. Models must be trained on team-collected or provided data
4.3. The model must operate in real time
4.4. Teams must be able to explain how their model works
4.5. Use of standard architectures (e.g., CNNs) is allowed
5. Prohibited Practices
5.1. Hardcoding the track (e.g., predefined movement without camera input)
5.2. Manual control during autonomous runs
5.3. Use of unauthorized sensors
5.4. Remote control from external systems (e.g., phone, laptop during run)
5.5. Copying code from other teams
5.6. Using pre-trained models without understanding them
5.7. Modifying the competition environment (e.g., moving track elements)
5.8. Interfering with other teams’ work
6. Run Rules
6.1. The better result is counted
6.2. The run starts on instructor’s signal
6.3. The run ends when:
- the robot finishes the track
- the time limit is exceeded
- the robot stops or gets stuck
6.4. The robot must operate autonomously when required
7. Timing and Penalties
7.1. Time is measured from start to finish
7.2. Penalties are added for errors
Penalties:
- Touching the robot: +5 seconds
- Collision with obstacle: +3 seconds
- Leaving the track: restart
- Failure to finish: DNF (0 points for the round)
8. Testing Conditions
8.1. All teams compete under the same conditions
8.2. The track is identical for all participants
8.3. Lighting conditions may change during competitions
8.4. The instructor may introduce minor track modifications
8.5. Teams do not have prior access to the final track
9. Scoring
9.1. Points are awarded based on competition results
9.2. Solution quality is also evaluated
Evaluation components:
9.3. Competition results – 40%
9.4. Documentation and Solution quality (mode, code on github required) – 30%
9.5. Presentation – 20%
9.6. Reproducibility – 10%
10. Documentation
10.1. Each team must submit a report
10.2. The report must include:
- approach description
- dataset description
- model architecture
- error analysis
- conclusions
10.3. Reports must be submitted on time
10.4. Missing report results in grade reduction
11. Knowledge Verification
11.1. Teams may be asked to present their solution
11.2. The instructor may ask technical questions
11.3. Each team member should understand the solution
11.4. Lack of understanding may result in individual grade reduction
12. Reproducibility
12.1. The solution must be runnable by the instructor
12.2. Code must be organized and documented
12.3. A clear run instruction must be provided
12.4. Failure to reproduce results leads to point loss
13. Safety
13.1. Actions that may damage hardware are prohibited
13.2. The robot must be used as intended
13.3. Work must stop in case of malfunction
13.4. Teams are responsible for improper usage
14. Final Provisions
14.1. Final interpretation of the rules belongs to the instructor
14.2. Minor changes to the rules may be introduced during the course
14.3. The goal is learning, not only competition
14.4. Participation implies acceptance of these rules