To provide a platform for students and professionals in India to participate in global pre-standards activity and Build-A-Thon challenge related to AI/ML in 5G. IIT Delhi with support from India EU ICT Standardization Project, ITU and TSDSI ,launched the India specific Build-A-Thon aligned with ITU global challenge.

• Build upon the previous work done by IIT Delhi in the area of focus group machine learning in 5G and extend for solving practical problems in the network and society.
• Proof of concept development, culminating in demos.
• Collaboratively create knowledge base and publish in global publications.
• Prepare Final technical report.
• A knowledge dissemination summary and technical report would be submitted as technical.
• Report to SDOs to be treated as pre-standards input. This can lead to further standard proposals after due study and gap analysis of existing standards.

Prizes:

• The winning team will get opportunity to enter the global ITU 2022 Build-A-Thon challenge.
• Study tour to EU for winning team.
• Winning team (with most practical design) will be invited to IITD for testing their solution on actual hardware.

Winners:

---

#	Team Name	Position
1	Roger That	Winner
2	AI ML SSD	Ist Runners Up
3	TCS Smart Machine	IInd Runners Up

---

Important Dates:

---

#	Date	Mode of Event	Event Title
1	12th Sep, 2022	Online	Round Table (IITD)
2	23rd Sep, 2022	Hybrid(Online/Offline)	Round Table 2 (IITD)
2	Deadline: 30th Sep, 2022	Online	Pre-evaluation of Problem Statement II (IITD)
2	Deadline: 3rd Oct, 2022	Online	Pre-evaluation of Problem Statement I (IITD)
2	11th Oct, 2022	Online	Final Submissions of Solution (IITD)
3	14th Oct, 2022	Online	Declaration of Results (IITD)
4	14th Oct, 2022	Hybrid	Build-a-thon Workshop, with invited talks from EU Experts (IITD)
5	28th Oct, 2022	Online	Start of Global round (ITU)
6	25th Nov, 2022	Online	Announcement of Global Awards (ITU)
7	Dec, 2022		EU trip to winners

---

Past Events

• Expert Talks , Recorded Session ,Bharti School, 2:00-4:30 pm 14^th Oct, 2022. Click here to view...
• Round Table (12^th Sep, 2022): Recorded Event Link: https://youtu.be/UedZYsrTPQg
• August Workshop:-Build-A-Thon: AI/ML driven Low Latency Closed Loop Control  https://youtu.be/pHqNHQxiBEo
• July Webinar: Build-A-Thon: AI/ML driven Low Latency Closed Loop Control https://youtu.be/Xmzc2HskQik

Pre-Evaluation

All the participants are requested to submit their prediction result for pre-evaluation: For Problem Statement I : Slip Detection and Force Estimation-Deadline is 3rd Oct, 2022, 11:59:59 pm Indian Standard Time For Problem Statement II : Object Detection - Deadline is 30th Sep, 2022, 11:59:59 pm Indian Standard Time Please follow Instruction set mentioned below for each Problem Statement. Please stick to the deadline. There will be no extension of the deadline. The results will be reported in a couple of days. This activity has a weightage of 20% and counted towards final results. It is an important activity and will help the teams to evaluate their performance and their respective positioning in the contest. Hence all teams are requested to submit for pre-evaluation. A Leaderboard ranking will be posted on the webpage for the pre-evaluation results. Teams may subsequently optimize their models before the final submission on 11th Oct, 2022.




Problem statement I - Slip Dtetection and Force Estimation

• Training Dataset Download
• Sample Dataset download
# Data format: Timestamp,J0_position,J0_Force,J1_position,J1_Force,J2_position,J2_Force,J3_position,J3_Force,J4_position,J4_Force,J5_position,J5_Force,J6_position,J6_Force,J7_position,J7_Force,J8_position,J8_Force,J9_position,J9_Force,J10_position,J10_Force,J11_position,J11_Force,J12_position,J12_Force,J13_position,J13_Force,J14_position,J14_Force,J15_position,J15_Force,Size,Weight_kG,SLIP,CRUMPLE # Metadata: 1. Ji_pos, Ji_Force denote the angular configuration and force experienced on joint Ji 2. Size denotes the shape of the object (in cm). 3. Weight_kG denotes the weight of the object (in kg). 4. Slip ∈{0,1} denotes whether the object is slipping [a value of 1 denotes slipping and a value of 0 denotes not-slipping] 5. CRUMPLE ∈{0,1} denotes whether or not the object has deformed [a value of 0 denotes not-crumpling and a value of 1 denotes crumpling]




Problem statement II - Object Detection

• Training Dataset download
  
• Sample Dataset download
# Data format: Timestamp,J0_position,J0_Force,J1_position,J1_Force,J2_position,J2_Force,J3_position,J3_Force,J4_position,J4_Force,J5_position,J5_Force,J6_position,J6_Force,J7_position,J7_Force,J8_position,J8_Force,J9_position,J9_Force,J10_position,J10_Force,J11_position,J11_Force,J12_position,J12_Force,J13_position,J13_Force,J14_position,J14_Force,J15_position,J15_Force,Weight_kG,Object_Held # Metadata : 1. Ji_pos, Ji_Force denote the angular configuration and force experienced on joint Ji 2. Weight_kG denotes the weight of the object (in kg). 3. Object_Held denotes the label of the object.





Final Submission Instructions:

FINAL SUBMISSION INSTRUCTIONS: ########################################################### 1. Each submission must be in ZIP format. The name of the ZIP file must be the name of the team/group (only ASCII characters, spaces, without any special characters, or symbols). 2. Upon unzipping, there should be a folder with your groupname (only ASCII characters, spaces, without any special characters, or symbols). 3. Inside the folder, there should be the following files (exactly as named below): - task1.py, this script contains the code that reads the test set and predicts the specified output for Task Problem Statement I: Slip Detection and Force Estimation in CSV format (one per line). The format of test set will be same as pre-evaluation dataset (i.e., the path will be the root folder containing all CSV files in the testset) - task2.py, this script contains the code that reads the test set and predicts the specified output for Task Problem Statement II: Object Detection in CSV format (one per line). A sample output is here. The format of test set will be same as pre-evaluation dataset (i.e., the path will be the complete path to ONE AND ONLY CSV file in the testset) - requirements.txt, this contains the requirements to load the dependent modules for your scripts. - task1_training.py, this script contains the code to read a train set and train a model with the data that must be saved/dumped as ./model_checkpoint_task1 - task2_training.py, this script contains the code to read a train set and train a model with the data that must be saved/dumped as ./model_checkpoint_task2 - model_checkpoint1 - model_checkpoint2 An example submission ZIP (named as "TuringGroup.zip") is linked below the submission instructions. 4. If you have worked on modelling of data on only one of the problem statements, then the code for the other problem need not be there. For example, if you have worked on the problem of Object Detection (i.e., Problem Statement 2), then the files task1.py, task1_training.py and model_checkpoint1 need not be there. 5. The code submitted must be written in Python 3.7. We will run an automatic script to run your submissions in the following format in the following phases: 5.a) First we will run "pip install requirements.txt" that will install all the dependencies needed for your code to run. 5.b) Then For task1 (Problem Statement I), we will run python task1.py --model_path ./model_checkpoint1 --input_data /path/to/testset/directory --output ./path/to/predictions/directory --input_data argument contains the path to the directory of the hidden test set. --input_data expects the path of a root directory where multiple CSV files exist belonging to the test-set. Your script must parse this argument and be able to read all CSV files from this directory path. The format of the input is CSV (exactly the same as the pre-evaluation dataset). --output argument defines the path to the directory where the predictions are to be dumped. Your scripts must parse this argument and be able to dump/save the output predictions (in CSV format) in the directory (specified by --output argument), corresponding to each input CSV file (with the same filename as that of the input CSV file). Individual output CSV files will contain TWO LABELS (separated with a comma) per line corresponding to each row in the input CSV. The first line of all the output CSV files must exactly be as "Slip, Crumple" referring to the two labels. --model_path argument contains the path to the model checkpoint/dump that you have previously trained for this problem. python task1_training.py --model_path ./model_checkpoint1 --input_data ./path/to/trainset/directory --input_data argument contains the path to the directory of the training set. --input_data expects the path of a root directory containing multiple CSV files belonging to the training set. Your script must parse this argument and be able to read CSV files from this path. The format of the input is (exactly the same as training dataset) --model_path argument contains the path where the model checkpoint will be dumped. 5.c) Then for task2 (Problem Statement II), we will run python task2.py --model_path ./model_checkpoint2 --input_data ./path/to/testset.csv --output ./predictions_task2.csv --input_data argument contains the end path to the hidden test set. --input_data expects the argument to be the complete path to the single CSV file representing the testset. Your script must parse this argument and be able to read input from this path. The format of the input is CSV (exactly the same as pre-evaluation dataset for this problem) --output argument defines the path where the predictions are to be dumped in CSV format. Your scripts must parse this argument and be able to save the output prediction in CSV format - one prediction per line. The output CSV file will contain only ONE LABEL per line corresponding to each row in the input CSV. The first line of the output CSV file must exactly be as "Object_Held", referring to the label of the object. --model_path argument contains the path to the model checkpoint/dump that you have previously trained for this problem. python task2_training.py --model_path ./model_checkpoint2 --input_data ./path/to/trainset.csv --input_data argument contains the path to the training set. Your script must parse this argument and be able to read input from this path. The format of the input is csv (exactly the same as training dataset) --model_path argument contains the path where the model checkpoint will be dumped. 6. If you have worked on only one problem statement, we will skip the code for the other task in 5.b or 5.c. 7. To be graded, your submission must contain code for at least one problem statement. GRADING INSTRUCTIONS: 1. Your code is auto graded. The code that does not run through steps defined in Point 5a) and either of 5b) and 5c), will lead to disqualification of the team that submitted that code. 2. The metrics of grading will depend on the following: - accuracy of the model (we will view the accuracy of the model on the output predictions from your code on the hidden test set). - time taken by the scripts/code takes to generate the predictions and/or perform training. - the size of the model checkpoint(s). The last date for final submission for both the Problem Statement is 11th October 2022, 11:59:59 pm Indian Standard Time. Example file for final submission: TuringGroup.zip





Important links

https://github.com/vrra/FGAN-Build-a-thon-2022





Top Three Leader Board Positions for Pre-evaluation

Leader board for pre-evaluation is presented here on the basis of "Accuracy" achieved and for any late submission 10% mark is deducted of the original score




Problem Statement I: Slip detection and Force Estimation

---

#	Team	Position
1	TCS Smart Machine	Ist
2	AI ML SSD	IInd
3	Roger That	IIIrd

---

Problem Statement II: Object Detection

---

#	Team	Position
1	Roger That	Ist
2	AI ML SSD	IInd
2	TCS Smart Machine	IIIrd

---