Signed in as:
Signed in as:
Technological singularity is a "hypothetical point in time at which technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization." Ref: Wikipedia.
The AI/ML ladder, as shown, is a unique SAAR vision of technological singularly, starting at a low lever of motion ability and ending up at a top, limitless, artificial creative thinking ability.
In between these two extreme layers the AI ladder goes upwards with sensing and artificial learning ability. It then continues to climb to decision making ability under uncertain environment and decision optimization ability with conflicting objectives, which are common goals of SAAR AI/ML projects.
The AI ladder graph also reminds us the level of interdisciplinary Human Resource which is needed to run a successful AI/ML activity in an academic or business environment. This includes - mechanical electrical and software engineers, information technology specialists, system analysts as well as business, marketing and financial managers to define and address target needs.
The required skill sets that any exoskeleton company which is interested in AI/ML technology, include - Industry 4.0 automation, IoT data generation, 5G communications, cloud and edge computation, possible global cashless Crypto currency transactions, and target market skillsets such as health care and rehabilitation in SAAR's case.
New AI/ML technology allows our business to differentiate our products with respect to competition with endless potential improvements. Our exoskeleton products are complex electromechanical robots. They are programmed to provide a desired supportive motion to handicapped users, subject to random (e.g. unexpected) and non linear (e.g. abrupt) user’s reactions.
There are many ways of controlling Robots. Some Examples include - bang/bang- for least energy, PID - the most common, fuzzy logic - for non linear systems, Kalman filters - for random systems and AI Reinforcement Learning (RL) which may combine features from all of the above.
RL, in exoskeletons, is based on a complex dynamic simulation in combination with an iterative process of a reward function optimization. The exoskeleton robot, in our case, is the environment. Its controller is a RL neural network (NN) with hundreds of parameters (not only 3 such as in PID), which are being tuned in an iterative, simulated, learning process. Once tuned they are deployed in the real system as the controller. The NN controller receives as an input multiple State signals from the environment sensors, such as position, velocity, acceleration, vibration, force, temperature, sound and vision. An Actor applies theses signals to the NN, with its hundreds of parameters, and outputs a set of control actions to the environment, such as optimal torque commands to the robot actuators, and safety signals to the users. The Action of the Actor changes the State of the environment. During the simulated RL process, a Critic provides the Actor a Reward (or penalty) for its action. Then a system Agent adjusts the NN parameters by minimizing the Bellman equation, such that the Critic’s and the Actor’s estimates of the total possible Reward of the entire motion process is closer to each other. The Reward function could be any combination of desired system performance, including those with conflicting interests, such as minimum error in target motion profile, maximum process speed, maximum safety, minimum consumed energy, maximum user comfort.
Our SAAR team has many years of experience in automation, system analysis, programming, mechatronics engineering, IT and motion control with successful record of state of the art products and business development which we are using now in this startup for helping handicapped people.