Internet of Things (IoT) is a new paradigm that integrates physical items from a variety of domains, such as human health, industrial processes, environmental monitoring and home automation with the Internet. In addition to many advantages, It creates security issues and expands the number of gadgets we use on a daily basis. The proposed deep learning-based intrusion detection methods test against a range of threats to determine their efficacy and offer suggestions for how well they perform in IoT intrusion detection. However, it is difficult to apply traditional Intrusion Detection system techniques to the Internet of Things due to its unique characteristics, including devices, specific protocol stacks, standards and limited resources. A new variational Autoencoder based Deep learning framework for Intrusion Detection (AUTODEEP-ID) has been proposed to address this problem and detect attacks in an Internet of Things environment. The suggested approach makes use of a BIGRU to categorize data into attacks and a Variational Autoencoder to extract pertinent features. The efficiency of the suggested approach is evaluated by recall, precision and accuracy. The observational findings shows that the AUTODEEP-ID detects DDOS and U2R as 0.3% and 0.2 % respectively.